Adaptive train model

Abstract

An adaptive train model (ATM) for predicting train consist reactions to specific stimuli using a system including at least one measurement sensor located on the train consist, a data base, and a computer. The ATM collects sensor data as the consist is moving, determines a consist force balance utilizing the sensor data and the computer, determines a set of consist coefficients using the computer, and predicts train consist kinetic characteristic values using the consist force balance and the set of consist coefficients.

Description

BACKGROUND OF THE INVENTION

This invention relates generally to predictive modeling, and more particularly to predicting train consist reactions to specific stimuli.

Typically, when a train consist moves from one location to another, knowledge of the effects of incurred stimuli of the consist as the consist moves along the tracks is useful for increasing the efficiency consist movement. Therefore, a train consist predictive model would aid railroad transportation companies in improving operations. However, modeling the movement of a train consist is complex and time consuming.

Therefore, it would be desirable to provide an effective method of modeling movement of a train consist so that kinetic characteristics, such as stopping distance, speed, and acceleration can be determined and utilized to improve the movement of freight from one location to another.

BRIEF SUMMARY OF THE INVENTION

In an exemplary embodiment, an adaptive train model (ATM) includes a semi-empirical mathematical model of a moving train consist used for simulating and predicting train reaction to external stimuli. The consist includes at least one locomotive and at least one railcar. The ATM utilizes a system including at least one sensor located on the train consist, a database, and a computer.

More particularly, the ATM is used for predicting such things as, train acceleration, train speed after one minute, and a shortest braking distance of the train consist. The model is adaptive because it is built and updated while the train is moving. The model has a set of unknown parameters that are updated as the train consist is moving and new data are collected. Then, based on the most recently measured data, those parameters are used for predicting movement characteristics of the train consist.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG. 1 is a schematic of a system used to implement an adaptive train model (ATM) in accordance with one embodiment of the present invention;

FIG. 2 is a schematic diagram of a compressed air system, with an auxiliary reservoir, used in a train consist utilizing the system shown in FIG. 1;

FIG. 3 is a chart depicting the difference between actual stopping distance and the predicted stopping distance determined utilizing the ATM with no initial values;

FIG. 4 is a chart depicting the stopping distances for a consist determined by the ATM using initial values previously collected;

FIG. 5 is a chart depicting the stopping distances for a consist determined by the ATM using information collected for the consist on a prior trip where the load was different from the present load;

FIG. 6 is a chart depicting variations of the predicted stopping distance for a consist determined by the ATM using as initialization information only some of the data collected based on values from a prior consist;

FIG. 7 is a chart depicting the predicted stopping distance for a consist determined by the ATM using modified initial data;

FIG. 8 is a chart depicting the predicted stopping distance determined by the ATM for the stopping distance for a consist as the brakes are fully applied instantly, using no initialization values;

FIG. 9 is a chart depicting variations of the predicted stopping distance determined with the ATM using initialization information based on data collected after three trips on the same track, and using the predictions made on the third trip as references; and

FIG. 10 is a chart depicting an error of an estimated safe stopping distance for a consist using the ATM initialized with weighted information.

DETAILED DESCRIPTION OF THE INVENTION

As used herein, the terms “train consist” and “consist” mean a train having at least one locomotive and at least one railcar. However, a “consist” may include a plurality of locomotives physically connected together with one locomotive designated as a lead locomotive and other locomotives designated as trailing locomotives, and a combination of railcars (freight, passenger, bulk) physically connected to the locomotives.

FIG. 1 is a schematic of a system 10 used to implement an adaptive train model in accordance with one embodiment of the present invention. System 10 includes at least one sensor 12 located on a railcar (not shown) of a train consist (not shown), and a computer 14. Computer 14 includes a database 16, a processor 18 capable of executing all functions of computer 14 and a memory 20 sufficient to hold data and algorithms used to implement the adaptive train model. Although system 10 is shown having only one sensor 12, system 10 may include a plurality of sensors 12. Sensor 12 monitors known train consist stimuli relating to moving train consist kinetics, such as, speed, pressure in consist brake pipe, consist position on a track, track configuration, rolling forces, aerodynamic forces, elevation caused forces, braking forces caused by direction changes, and braking forces caused by application of railcar brake shoes (not shown). Sensor 12 communicates stimuli data to computer 14, which uses processor 18 to interpret the stimuli data. Processor 18 utilizes a modeling algorithm, described below and hereinafter referred to as an adaptive train model (ATM), to process the stimuli data, thereby producing coefficients and operational data that are used simulate train consist kinetic characteristics and predict train consist movement. In one embodiment, the coefficients and operational data are stored in memory 20. In an alternative embodiment, the coefficients and operational data are stored in database 16 according the ATM.

The ATM is a semi-empirical model using a set of initially unknown coefficients. Those coefficients are determined and adjusted based on the most recent collected stimuli data and train reaction to those stimuli, such as acceleration. Once determined, the coefficients are used for predicting train kinetic characteristics for a period of time. The ATM is semi-empirical because it takes in account the most important elements of consist kinetics but considers as unknown their interaction. The ATM makes the assumption that the train consist length is constant, the load the consist is carrying is uniformly distributed along the consist, the brakes on all railcars have the same unknown efficiency, and the spiral curvature (1/R) of the track has a linear variation with distance.

Table 1, below, defines the notations that will be used to describe the ATM and how the notations are used to predict consist kinetic characteristics.

TABLE 1
Symbol	Meaning	Unit
a	Acceleration	feet/s2
a(t)	Acceleration as a function of time	feet/s2
Bi(t)	brake functions	feet/s2
Cl(t)	Braking effect caused by lateral friction when train is	feet
	in curve
Cp(t)	Braking effect caused by weight increase when train is	feet
	in curve
D	distance	feet
D(t)	dynamic brake	pounds
DC	degree of a curve (angle for 100 feet of track)1	degrees
Ei(t)	Elevation function	Feet
F	Force	pounds
g	Gravitational acceleration (9.82 m/s2 = 32.218 feet/s2)	Feet/s2
Ka	Corrective factor for the effect of aerodynamic friction	lbs/feet
Kbi	brake function coefficients	no unit
Kd	Corrective factor for the effect of dynamic brake	no unit
	application
Kei	Corrective factor for the effect of elevation change on	s−2
	segment i of the train
Kl	Corrective factor for the effect of lateral friction when	s−2
	train is in curve
Kp	Corrective factor for weight increase when train is in	s−2
	curve
Kr	Corrective factor for friction of a train rolling on	feet/s2
	straight horizontal track
Kri	release function coefficient	no unit
Krv	Dynamic corrective factor for friction of a train rolling	s−1
	on straight horizontal track
Kt	Corrective factor for the effect of throttle application	no unit
L	total train length	feet
Li	length of segment i	feet
lij	length of the segment i section j of the train	feet
M	total train mass	lbs
Mi	mass of segment i
mij	mass of the segment i section j of the train	lbs
Nax	Number of powered axles
p(t)	Pressure in brake pipe measured at front locomotive	psi
Pmax	Maximum pressure in brake pipe	psi
R	curve radius	feet
Ri(t)	release functions	feet/s2
L	train length	feet
T(t)	traction force	pounds
v	speed	feet/s
v(t)	speed as function of time	feet/s
vd	speed recorded in database	feet/s
W	total train weight	lbs
wij	weight of the segment i section j of the train	lbs
1The field CURVE in track database.

Model Description

The ATM determines the train kinetic elements based on a force balance. As the train moves, there are forces pulling the train forward and other forces pulling the backward. The forward forces accelerate the train movement, while the backward forces have a braking effect and create a deceleration. The effect of the vertical forces are considered only as an element affecting the rolling friction of the consist.

In an exemplary embodiment, a balance of forces applied to a consist at time t isF(t)=M(K_r+K_rvv(t))+K_av(t)²+MK_e1E₁(t)+MK_e2E₂(t)+MK_e3E₃(t)+MK_e4E₄(t)+MK_pC_p(t)+MK₁C₁(t)+K_b1B₁(t)+K_b2B₂(t)+K_b3B₃(t)+K_b4B₄(t)+K_r1R₁(t)+K_r2R₂(t)+K_r3R₃(t)+K_r4R₄(t)+K_dD(t)+K_tT(t).  (1)

The meaning of the function and coefficients is described below.

A rolling forces term, M (K_r+K_rv v(t)), is the sum of all forces that act on a train moving along an ideal straight horizontal line when no brake or throttle is applied. Those forces can be determined on each independent railcar. They are caused by the friction between rail and wheels, rail elastic deformation under railcar weight, deformation (compression) of the rail understructure caused by railcar weight, bearing friction, and other forces in direct relation with the railcar weight and railcar speed. Adding the effect of all railcars, gives a force that is dependent on train total mass and speed.

K_r is the only coefficient in the ATM that does not have an associated element. K_rv is associated with train speed. As the consist speed changes, K_rvv modifies the value of the rolling forces.

Aerodynamic forces are expressed by the term, K_a v(t)². The aerodynamic forces are caused by the friction between the moving train consist and the stationary air. The total force depends on locomotive aerodynamic profile, railcar aerodynamic profiles and the sequence of different railcar types along the consist, such as, flat bed, single stacked, double stacked, caboose, gondola, etc. For such reason the value of the coefficient should be different from one consist to another as each consist has its own aerodynamic profile.

Elevation caused forces are expressed by M (K_e1 E₁(t)+K_e2 E₂(t)+K_e3 E₃(t)+K_e4 E₄(t)). It is assumed that the consist has up to four segments, each uniformly loaded. This assumption is important for modeling the movement of very long consists, where distribution of weight has an important contribution on the speed determination. The size of each segment is determined from loading information provided in a consist description. If such information does not exist, all four segments are considered to be of equal length. For small consists (less than 20 railcars) or uniformly loaded consists, such as coal, ore or other specialized consists, the whole train is considered as one segment.

As a train travels over hills and valleys, some sections of the train are moving uphill, while other train sections are rolling downhill.

A force f_ij created by section j of segment i of the consist is tangential to the ground and has the valuef_ij=w_ijcos(π/2−α_ij).  (2)

Factor w_ij is the weight of consist section j of segment i. Since the load is assumed to be distributed uniformly along the segment, the weight of the section j of segment i should bew_ij=M_jgl_ij/L_j.  (3)

Then an elevation effect on consist section j of segment i isf_ij=M_igl_ij/L_icos(π/2−α_ij).  (4)

Values 1_ij are the “footage” determined by a track database using two successive elevation records. The first and the last sections of the consist should be determined according to the position of the head and the tail of the train with respect with the surrounding terrain. As the train travels, the size of the train sections are changing continuously, thus they are time dependent. Constraint

$\begin{array}{cc}{L}_i=\sum _{j=1}^n{l}_{\mathrm{ij}}\left(t\right)& \left(5\right)\end{array}$ is verified at each moment t. The value of cos(π/2−α_ij) is approximately the same as the value of the field “slope” in a track database divided by 100.

E_i(t) is determined by

$\begin{array}{cc}{E}_i\left(t\right)=\sum _{j=1}^n{l}_{\mathrm{ij}}\left(t\right){\mathrm{slope}}_{\mathrm{ij}}/100\mathrm{DB\_DIRECTION}.& \left(6\right)\end{array}$

The DB_DIRECTION factor defines the vertical direction each segment of the consist is moving. If the segment is on a downhill slope, DB_DIRECTION is set to negative one (−1), otherwise its value is one (1).

Term g M_i/L_i is a constant that was not included in (6). Thus, value of K_ei should be g M_i/L_I, givingK_ei=gM_i/L_i.  (7)

From equation (7)K_eiL_i/g=M_I, and  (8)

$\begin{array}{cc}\sum _{i=1}^4{K}_{\mathrm{ei}}{L}_i=\mathrm{Mg}=W.& \left(9\right)\end{array}$

Noise associated with the data collection and database inconsistencies are not considered, therefore, the sum is an approximation of train weight.

Braking forces caused by direction change are expressed by the term M (K_p C_p(t)+K₁ C₁(t)). Each change in track direction has three distinctive segments, the entry easement, the circular curve and the exit easement. The entry easement is a spiral with the curvature varying from 0 to a final value of 1/R. In other words, the radius of the turn varies along the easement from ∞ to R. The curve is a circle with a constant radius of R. The exit easement is symmetrical to the entry easement. It starts with radius R and ends with a radius equal with ∞. When a railcar is moving at constant speed along the circular section of the track, it is affected by the centrifugal acceleration determined byF_c=mv²/R.  (10)

When the railcar is moving along a straight track, 1/R is zero and the centrifugal force is nonexistent. The reason for building the entry end exit easements is to prevent the centrifugal force discontinuities from zero to the value created by the movement on circular track. Radius R is determined by using a D_C field from track database, givingR=180/π100/D_C=5729.58/D_C.  (11)

The centrifugal force is compensated for by raising the outer rail of the track, thereby tilting the railcar inward. The point of the railcar center of gravity O depends on the level of railcar loading and the type of load. There are two main forces acting on a railcar moving on a circular track, a railcar weight (w) and a centrifugal force (F_c). The centrifugal force is dependent on railcar mass, speed and track curvature. The railcar weight is dependent on railcar mass and gravitational acceleration. These two forces are composed into a resultant (w′). In order to prevent train derailment, the position of this force should always be between the two rails. Therefore,w′=sqrt(w²+F_c²), and  (12)w′=msqrt(g²+(v²/R)²)=msqrt((Rg)²+v⁴)/R.  (13)

The ideal position of this force is perpendicular to the track. Such position corresponds to a speed vd recorded in the data base for that particular curve and is strictly related with the amount of superelevation. The force w′ is decomposed in two components, one is perpendicular to the track (F_p), while the other is parallel to the track (F₁). The component perpendicular to the track is the “apparent” weight of the railcar and the component parallel to the track creates lateral friction. Therefore,F_p=w′cos(θ)=msqrt((Rg)²+v⁴)cos(θ)/R, and  (14)F₁=w′sin(θ)=msqrt((Rg)²+v⁴)sin(θ)/R.  (15)

The angle θ′ is between forces w and w′, givingtan(θ)=F_c/w=(mv²/R)/(mg)=v²/(Rg).  (16)

Θ is the superelevation angle and should be equal to the angle between forces F_p and w. The superelevation angle should satisfy the relationtan(Θ)=vd²/(Rg),  (17)sin(Θ)=vd²/sqrt((Rg)²+vd⁴);cos(Θ))=Rg/sqrt((Rg)²+vd⁴), and  (18)sin(θ)=v²/sqrt((Rg)²+v⁴);cos(θ)=Rg/sqrt((Rg)²+v⁴).  (19)

The angle θ is the difference θ′−Θ. Therefore,cos(θ)=cos(θ′−Θ)=cos(θ′) cos(Θ)+sin(θ′) sin(Θ), and  (20)sin(θ)=sin(θ′−Θ)=sin(θ′)cos(Θ)−cos(θ′)sin(Θ).  (21)

Which givescos(θ)=((Rg)²+v²vd²)/(sqrt((Rg)²+vd⁴)sqrt((Rg)²+v⁴)), and  (22)sin(θ)=Rg(v²−vd²)/(sqrt((Rg)²+vd⁴)sqrt((Rg)²+v⁴)).  (23)

Replacing the values in (14) and (15), givesF_p=m((Rg)²+v²vd²)/(Rsqrt((Rg)²+vd⁴)), and  (24)F₁=mg(v²−vd²)/sqrt((Rg)²+vd⁴).  (25)

These two forces are associated to the sections of track that are not tangent (straight). F_p represents the braking force caused by the effect of gravitation and centrifugal acceleration. If the effect of gravitation was already included in rolling forces, the relation in (24) should be adjusted by subtracting m g, givingF_p=m(((Rg)²+v²vd²)/(Rsqrt((Rg)²+vd⁴))−g).  (26)

The effect of all sections of the train that are on a circular track at a moment t is

$\begin{array}{cc}{F}_p=\sum _{i=1}^n{m}_i\left(\left({\left(R_{i}g\right)}^2+v^2{\mathrm{vd}}_i^2\right)/\left(R_i\mathrm{sqrt}\left({\left(R_{i}g\right)}^2+{\mathrm{vd}}_i^4\right)\right)-g\right),& \left(27\right)\end{array}$ and using (15) gives

$\begin{array}{cc}{F}_1=\sum _{i=1}^n{m}_{i}g\left(v^2-{\mathrm{vd}}_i^2\right)/\mathrm{sqrt}\left({\left(R_{i}g\right)}^2+{\mathrm{vd}}_i^4\right).& \left(28\right)\end{array}$

Value m_i is the mass of the train on each circular section i of track. For a uniform train loading the value should be proportional with the section length. The final expression of forces is then given by

$\begin{array}{cc}{F}_p=M/L\sum _{i=1}^n{l}_i(\left({\left(R_{i}g\right)}^2+v^2{\mathrm{vd}}_i^2\right)/\left(R_i\mathrm{sqrt}\left({\left(R_{i}g\right)}^2+{\mathrm{vd}}_i^4\right)\right)-g,\mathrm{and}& \left(29\right)\\ F_1=M/L\sum _{i=1}^n{l}_{i}g\left(v^2-{\mathrm{vd}}_i^2\right)/\mathrm{sqrt}\left({\left(R_{i}g\right)}^2+{\mathrm{vd}}_i^4\right).& \left(30\right)\end{array}$

Similar formulae are associated with forces developed by the train sections in easements. Since the radius of easement is variable, forces associated with easements are considered to be half of the forces created in circular sections. Such a supposition is correct if easements have a linear variation of a curvature 1/R. Removing constant terms, the expression of the “perpendicular force effect” for sections of train located in curve sections of track is

$\begin{array}{cc}{C}_p\left(t\right)=\sum _{i=1}^n{l}_i\left(t\right){\delta }_i\left(\left({\left(R_{i}g\right)}^2+v^2{\mathrm{vd}}_i^2\right)/\left(R_i\mathrm{sqrt}\left({\left(R_{i}g\right)}^2+{\mathrm{vd}}_i^4\right)\right)-g\right),& \left(31\right)\end{array}$ and the “lateral force effect” is

$\begin{array}{cc}{C}_1\left(t\right)=\sum _{i=1}^n{l}_i\left(t\right){\delta }_i\left(v^2-{\mathrm{vd}}_i^2\right)/\mathrm{sqrt}\left({\left(R_{i}g\right)}^2+{\mathrm{vd}}_i^4\right)).& \left(32\right)\end{array}$ where δ_i are 1 for circular sections and 0.5 for easements.

Braking forces caused by the operation of the consist brakes are expressed by K_b1 B₁(t)+K_b2 B₂(t)+K_b3 B₃(t)+K_b4 B₄(t). The sum of four terms K_bi B_i(t) denote the total braking force applied to the train. The value of this sum will differ from zero only when the brake shoes are pressing the wheels on at least one railcar. After an engineer reduces the pressure in brake pipe, the brake shoes along the consist are successively applied against the wheels increasing the total braking force to a maximum value, railcar by railcar. Although all shoes are pressing the wheels with the maximum force, the total braking force does not remain constant. As the train speed decreases, the friction coefficient between the wheels and shoes changes its value modifying the value of the train total braking force.

The friction coefficient between the wheels and shoes is in direct relation to the material composition of the brake shoes. Brake shoes can be made from cast iron, cast iron with high phosphorous, or high friction composition blend of organic and inorganic materials. In addition, the static and dynamic friction coefficients are different from one shoe type to another. The following formulae determine the friction coefficient.

For cast iron shoes the friction coefficient is,C_i=0.18+0.28*exp(−0.25*v), and  (33)

For high composition shoes the friction coefficient is,C_c=0.255+0.11*exp(−0.07*v).  (34)

FIG. 2 is a schematic diagram of a compressed air system 30 for a train consist using an auxiliary reservoir. Compressed air system 30 includes a consist brake pipe 32 that runs the length of the train consist, a main reservoir 34 for storing pressurized air, an auxiliary reservoir 36 for storing auxiliary pressurized air, a brake cylinder 38 that engages and disengages the brake shoes of a railcar (not shown), and an exhaust 40 that allows air to be vented from system 30. Although system 30 is shown having only one brake cylinder 38, each wheel of a railcar is coupled with a brake cylinder 38 such that a respective set of brake shoes are operated for each wheel. Brake application is caused by a reduction of air pressure in consist brake pipe 32. The dragging force created by the brake application is equal to the product between a friction coefficient C_f(v) and a force f_b(t) applied by the brake shoes to the wheels of each railcar, as the pressure builds up in brake cylinders 38. The value of the pressure in brake cylinders 38 relates to a pressure drop in brake pipe 32 Δp and an initial pressure in the auxiliary reservoir 36 Pa.

The variation of pressure in brake cylinders 38 depends on a variety of factors, such as, the length of brake pipe 32, air temperature, humidity, the operational status of a brake valve 42, and the amount of slack in the brake shoes. However, there are likely more factors, some unmeasurable or unknown, that effect the pressure in brake cylinder 38. To compensate for variations two models that evaluate the brake cylinder force in two opposite conditions are used, one supposes that the pressure is building fast in brake cylinders 38, while the other considers that the pressure is building much slower. Although, the two pressure models are designed to match the braking force developed by short trains and long trains, their linear combination can mimic the time variation of the braking force for a large variety of trains in many external conditions. The proper combination of those two pressure models with the friction coefficients is a critical task of the Adaptive Train Model.

The model for fast building pressure uses a pressure drop causing equalizationEp=(Pa^1.18285856+84.14968969)/12.43010076,  (35)a reference pressureRp=min(Ep,Δp), anda rigging equationRigg=min(0,max(1,(12.4301008*Rp−84.1496897)/Pa^1.18285856)).  (36)

The rigging value should be between 0 and 1. Therefore, brake cylinder maximum force equalsBc_f=(345.0445077*Rp−4986.482562)*Rigg, and  (37)modified time equalsT=max(0,(t−2.53325611)*(297.383526/(0.08709894*L+1))^0.77418566*[1/(−0.02072556*Pa−0.00036791*Pa²+6.47422981)+1/(−0.00423106*Rp+0.00137952*Rp²+0.02506324)].  (38)

If Δp>Ep, the modified time is replaced byT=T^{(1+25.2474268(Rp−Ep)/Pa−148.737743((Rp−Ep)/Pa)2)}.  (39)

Therefore, the braking force for fast building pressure is given byBf_f=min(0,max(1,(T+3.86950758*T²+0.23164628*T³)/(16367.9101+111.652789*T+27.61345048*T²−0.0026229*T³)))Bc_f.  (40)

The model for slow building pressure uses a pressure drop causing equalizationEp=(Pa^1.186922636+83.5394856)/12.47984,  (41)a reference pressureRp=min(Ep,Δp), and  (42)a rigging equationRigg=min(0,max(1,(12.47984*Rp−83.5394856)/P_a^1.18692264)).  (43)

The rigging value should be between 0 and 1. Therefore, brake cylinder force is given byBc_s=(346.46923*Rp−5015.0018)*Rigg, and  (44)modified time equalsT=max(0,(t−4.99998034)*[297.383526/(0.08709894*L+1)]^0.77418566*[1/(−0.00781443*Pa−0.00034658*Pa²+6.76259649)+1/(0.07532957*Rp+0.0020505*Rp²+0.93285061)].  (45)

If Δp>Ep, the modified time is replaced byT_s=T_s^{(1+2.47330639(Rp−Ep)/Pa−8.92733905((Rp−Ep)/Pa)2)}.  (46)

Therefore, the braking force for slow building pressure is given byBf_s=min(0,max(1,(T_s+2.00986206*T_s²+0.81412194*T_s³)/(0.00067603+169.361303*T_s+8.95254599*T_s²+0.58477705*T_s³)))Bc_s.  (47)

Taking into consideration the model for fast building pressure and the model for slow building pressure, the total braking force has four components

• • B₁=C_iBf_f,
  • B₂=C_cBf_f,
  • B₃=C_iBf_s, and
  • B₄=C_cBf_s.

The balance of forces equation (1) also takes into consideration the brake release pressures. When an engineer restores the pressure in the brake pipe to the initial value, the brake shoes are removed from wheels and the dragging force created by the brake application decreases. The ATM considers brake release as a force opposing the braking force. The ATM has to identify the proper coefficients such that the release force cancels the brake force when shoes are in fact completely removed from wheels. As in the brake application, two conditions are considered, a fast release model and a slow release model.

The model for fast release utilizes a pressure drop causing equalizationEp=(Pa^1.18285856+84.14968969)/12.43010076,  (48)a reference pressureRp=min(Ep,Δp), and  (49)a rigging equationRigg=min(0,max(1,(12.4301008*Rp−84.1496897)/P_a^1.18285856)).  (50)

The rigging value should be between 0 and 1. Therefore, brake cylinder maximum reference force is given byBc_f=(345.0445077*Rp−4986.482562)*Rigg.  (51)

Thus, the fast release force is given byRf_f=min(0,max(1,(t+3.86950758*t²+0.23164628*t³)/(16367.9101+111.652789*t+27.61345048*t²−0.0026229*t³)))Bc_f.  (52)

The model for slow release utilizes a pressure drop causing equalization:Ep=(Pa^1.186922636+83.5394856)/12.47984,  (53)a reference pressure:Rp=min(Ep,Δp), anda rigging equation:Rigg=min(0,max(1,(12.47984*Rp−83.5394856)/P_a^1.18692264)).  (54)

The rigging value should be between 0 and 1. Therefore brake cylinder reference force:Bc_s=(346.46923*Rp−5015.0018)*Rigg.  (55)

Thus the slow release force is given by:Rf_s=min(0,max(1,(t+2.00986206*t²+0.81412194*t³)/(0.00067603+169.361303*t+8.95254599*t²+0.58477705*t³)))Bc_s.  (56)

Taking into consideration the model for fast release and the model for slow release, the total release force has four components

• • R₁=C_iRf_f,
  • R₂=C_cRf_f,
  • R₃=C_iRf_s, and
  • R₄=C_cRf_s.

The dynamic brake force is represented by K_d D(t). The value in pounds of D(t) is determined byD(t)=I²v²N_ax/591.43,  (57)where I is a current in Amperes, v is a speed in feet/s and N_ax is a number of powered axles.

The traction force is represented by K_t T(t), where T(t) is the traction effort and is dependent on locomotive type and train speed. K_t will be one if the train has only one locomotive, the locomotive efficiency is 100% and the locomotive wheels do not slip.

For trains using more than one locomotive type, or not using the same number of locomotives all the time, T(t) will be the sum of traction efforts developed by all active locomotives.

Since a train mass M does not change value during the trip, the relation in (1) is divided by the total train mass. The result is the expression of train accelerationa(t)=K_r+K_rvv+K_av²+K_e1E₁(t)+K_e2E₂(t)+K_e3E₃(t)+K_e4E₄(t)+K_pC_p(t)+K₁C₁(t)+K_b1B₁(t)+K_b2B₂(t)+K_b3B₃(t)+K_b4B₄(t)+K_r1R₁(t)+K_r2R₂(t)+K_r3R₃(t)+K_r4R₄(t)+K_dD(t)+K_tT(t).  (58)

In this relation all current and past values of speed, acceleration, throttle position, pressure in brake pipe, and position on track can be used for determining coefficients K_i. Once the coefficients are determined, they can be used for predicting the consist speed, acceleration, and other values at any moment of time.

Determining the Coefficients.

In one embodiment, to determine the coefficients for the ATM, the Least Squares Method (LSM) is utilized. In one embodiment, the behavior of a real world object depends on a set of measurable independent stimuli x₁, x₂, . . . . x_n. To any set of stimuli, the object provides a measurable reaction y. Therefore a mathematical relation that approximates the reaction of the object to those stimuli is developed.

The independent set of variables x (stimuli) and the dependent variable y (reaction), are measured. Measuring those variables, m times gives the following table of measurements M:X_1,1 X_1,2 . . . X_1,a Y₁

$\hspace{1em}\begin{array}{ccccc}{x}_{2,1}& x_{2,2}& \dots & x_{2,n}& y_2\\ \dots & \dots & \dots & \dots & \dots \\ x_{m,1}& x_{m,2}& \dots & x_{m,n}& y_m.\end{array}$

The dependent variable y can be determined by a linear combination of other independent variables x, giving the equation

$\begin{array}{cc}y=\sum _{j=1}^n{a}_j{x}_j.& \left(59\right)\end{array}$

The dimension m of the table is variable. At any time another experiment can be made and the new set of values added to a previously built table.

For any set of chosen or determined values of a_j coefficients an approximation y′ is determined. The difference between the measured y and the determined y′ is the approximation error ε, where

$\begin{array}{cc}{\varepsilon }_i=y_i\underset{i=1}{\overset{n}{-\sum }}{a}_j{x}_{i,j}\mathrm{for}i=1,2,\dots ,n.& \left(60\right)\end{array}$

By choosing different sets of a_j coefficients, the errors ε_i change. In one embodiment, the values of a_j that minimize the expression

$\begin{array}{cc}E=\sum _{i=1}^m{\varepsilon }_i^2=\sum _{i=1}^m{\left(y_i\underset{j=1}{\overset{n}{-\sum }}{a}_j{x}_{i,j}\right)}^2& \left(61\right)\end{array}$ are calculated. The global error E depends on a_j values only, while all other elements in relation (61) are measured numerical values. Then the partial derivative of E with any a_k will be zero, for example,∂E/∂a_k=0 for k=1, 2, . . . n.  (62)

Therefore, replacing the value of E from (61) in (62) and differentiating gives the equation

$\begin{array}{cc}-2\sum _{i=1}^m\left(\left(y_i-\sum _{j=1}^n{a}_j{x}_{i,j}\right)\sum _{j=1}^n\left(\partial a_j/\partial a_k{x}_{i,j}\right)\right)=0\mathrm{for}k=1,2,\dots n.& \left(63\right)\end{array}$

The term a ∂ a_j/∂ a_k is 1 when j is equal with k and zero in all other cases. Ignoring the constant term −2 gives the equation

$\begin{array}{cc}\sum _{i=1}^m\left(\left(y_i-\sum _{j=1}^n{a}_j{x}_{i,j}\right)x_{i,k}\right)=0\mathrm{for}k=1,2,\dots n.& \left(64\right)\end{array}$

Changing the summation order and moving known elements on the right hand of the equation gives

$\begin{array}{cc}\sum _{j=1}^n{a}_j\left(\sum _{i=1}^m{x}_{i,j}{x}_{i,k}\right)=\sum _{i=1}^m{y}_i{x}_{i,k}\mathrm{for}k=1,2,\dots n.& \left(65\right)\end{array}$

Both summations by i are using measured numbers. Those summations provide the matrix that is used for determining the unknown a_i coefficients. Using the notations

$\begin{array}{cc}{X}_{k,j}=X_{j,k}=\sum _{i=1}^m{x}_{i,j}{x}_{i,k},\mathrm{and}{Y}_k=\sum _{i=1}^m{y}_i{x}_{i,k},& \left(66\right)\end{array}$ gives the final form of the system of equations

$\begin{array}{cc}\sum _{j=1}^n{X}_{k,j}{a}_j=Y_k\mathrm{for}k=1,2,\dots n.& \left(67\right)\end{array}$

The relation (67) is a system of n equations with n unknown variables a_j. The matrix of this system X and the right hand term Y are determined from the measured data. The values of unknown coefficients a_j are found by solving the system of equations (67). After determining coefficients a_j, the individual error for each measurement is determined using

$\begin{array}{cc}{\varepsilon }_i=y_i-\sum _{j=1}^n{a}_j{x}_{i,j}\mathrm{for}i=1,2,\dots m.& \left(68\right)\end{array}$

Replacing the expression of error e_i in (65) gives

$\begin{array}{cc}\sum _{i=1}^m{\varepsilon }_i{x}_{i,k}=0\mathrm{for}k=1,2,\dots n.& \left(69\right)\end{array}$

If one of the x_i,j has the value one, then the individual errors also have the property

$\begin{array}{cc}\sum _{i=1}^m{\varepsilon }_i=0.& \left(70\right)\end{array}$

If measurements are not all made in the same conditions, weighting factors reflecting our confidence in the validity of each measurement are used. In such a case, the expression is minimized by

$\begin{array}{cc}E=\sum _{i=1}^m{w}_i{\varepsilon }_i^2=\sum _{i=1}^m{w_i\left(y_i-\sum _{j=1}^n{a}_j{x}_{i,j}\right)}^2,& \left(71\right)\end{array}$ where each independent error has a different weight. Additionally,

$\begin{array}{cc}{X}_{k,j}=X_{j,k}=\sum _{i=1}^m{w}_i{x}_{i,j}{x}_{i,k},\mathrm{and}{Y}_k=\sum _{i=1}^m{w}_i{y}_i{x}_{i,k}.& \left(72\right)\end{array}$

If the approximation has a free term, the property

$\begin{array}{cc}\sum _{i=1}^m{w}_i{\varepsilon }_i=0\mathrm{applies}.& \left(73\right)\end{array}$

The relation (58) defines the train acceleration. The following table is a mapping of the train movement model to least square method notation.

Train LSM
Kr    A0
l     X0
Krv   A1
v     X1
Ka    A2
v2    X2
Ke1   A3
E1(t) x3
Ke2   a4
E2(t) x4
Ke3   a5
E3(t) x5
Ke4   a6
E4(t) x6
Kp    a7
Cp(t) x7
Kl    a8
Cl(t) x8
Kb1   a9
B1(t) x9
Kb2   a10
B2(t) x10
Kb3   a11
B3(t) x11
Kb4   a12
B4(t) x12
Kr1   a13
R1(t) x13
Kr2   a14
R2(t) x14
Kr3   a15
R3(t) x15
Kr4   a16
R4(t) x16
Kb9   a14
R1(t) x14
Kt    a17
T(t)  x17
acc(t)

The measured parameter x₀ always has the value 1. Time t is not explicitly used in this model. All x_i values are determined using the latest measured acceleration, speed, train position and pressure in equalizing reservoir 36 shown in FIG. 2.

The X matrix is symmetrical with the positive diagonal. As a result, indefinitely adding elements to the matrix causes data degeneration and loss of precision. In order to prevent data degeneration each measurement is weighted. Thus, the weight of the present measurement is set to one, the weight of the measurement before that is set to a smaller value and the weight of each preceding measurement is set to yet a smaller value.

The value ε is set to a positive number smaller than 1 and the series 1, (1−ε), (1−ε)², (1−ε)³, . . . converges to zero. By considering the series of weights for consecutive measurements, after an infinite number of measurements, any positive term on the diagonal of the LSM matrix will have a value limited according to the equation,

$\begin{array}{cc}x\left[i,i\right]\le 2^{}\sum _{k=0}^{\infty }{\left(1-\varepsilon \right)}^k={}^2\left(1-{\left(1-\varepsilon \right)}^{\infty }\right)\text{/}\left(1-\left(1-\varepsilon \right)\right)=2^{}\text{/}\varepsilon ,& \left(74\right)\end{array}$ where ℑ is the maximum value between one and the largest absolute value of the considered parameter. From (74), applying the weighting procedure presented above to LSM generates a matrix that has all terms on the diagonal smaller than ℑ²/ε, where 0<ε<1. Additionally, equally weighted measurements (corresponding to ε=0) create a condition for floating point errors, and that large values of ε (almost one) assure better precision for number representation, but gives too small of a weight to prior samples, practically eliminating them from selection.

The optimal value will be somewhere between zero and one. Using ε=10⁻³ with 32 bit floating point representation provides 4 correct decimal digits. Using ε=10⁻⁵ with 64 bit floating point representation provides 11 correct decimal digits of the result.

The term x[1,1] is associated to the free term. Without any weighting, this number is equal with the number of samples considered for determination. When using weights, the term is called “the apparent number of samples”. Using the weighting procedure the value of x[1,1] converges to 1/ε.

The sum in (74) is performed for an infinite number of terms and there is no need to store more data than the current evaluated x vector. The weighting of data is realized by “aging” the X matrix and the Y vector before adding the latest collected data. If X(n) is the matrix X for step n and Y(n) is the vector Y in the same step, the next step elements areX(n+1)=(1−ε)X(n)+∥x_n+1,jx_n+1,k∥ for j,k=1, 2, 3, . . . m, and Y(n+1)=(1−ε)Y(n)+y_n+1∥x_n+1,k∥ for k+1, 2, 3, . . . m.  (75)

From (75), the following corollary is derived.If X(n) has the inverse P(n) and X(n+1)=X(n)+x(n+1)x(n+1)^T, then P(n+1)=P(n)−[P(n)x(n+1)][[x(n+1)^TP(n)]/(x(n+1)^TP(n)x(n+1)+1)].  (76)

However, if there is problem in the initial matrix, X(0) is null and it has no inverse. To overcome this inconvenience without destroying the precision the initial value of X(0) is set to some small values on the diagonal, for example 10⁻⁵, with the rest of the matrix set to zero. During the process the small value is overwritten by larger x(n)x(n)^T values. The inverse of X(0) is P(0) with all elements set to zero except for the diagonal that has the inverse of each element from the diagonal of X(0) divided by m.

A Y vector is updated according toY(n+1)=Y(n)+accx(n+1), and  (77)the values of a_j coefficients are determined according toa(n)=P(n)Y(n).  (78)

When performing the matrix operations the total number of computer operations is 13 m²+2 m+1. The Gauss-Jordan triangulation method, requires for solving the same problem, m³+(m²−m)/2 operations.

Model Confidence

All variables in the ATM are independent. Therefore, all coefficients determined will be independent. Thus, the real train behavior and measurements are not affected by large errors, and after a short period of time, the value of each coefficient will be approximately the same from one second to another.

The stability of coefficient values effects the stability of predictions that are provided. One way to measure the quality of information gathered by the model is to monitor the differences between the predicted speed and the measured speed. The speed may be predicted for one minute, 15 seconds, or only one second. The potential disadvantage is that the quality of a prediction at a time after the prediction was actually made is unknown.

For a one second speed prediction the response is available after one second, but is affected by the measurement noise, and the one minute prediction provides a response after one minute, which is too late. A five second or fifteen second prediction is sufficient for judging the quality of model predictions.

Each speed measurement v_k is associated with the value {overscore (v)}_k, predicted a few seconds earlier. For a large set of measurements or predictions the variance σ is associated according to the equation,

$\begin{array}{cc}{\sigma }^2=\sum _{k=1}^n{\left({\stackrel{\_}{v}}_k-v_k\right)}^2/\left(n-1\right).& \left(79\right)\end{array}$

The only values that have to be stored are n, the sum of the square values

$\sum _{k=1}^n{\left(v_k\right)}^2,$ and the sum of the values

$\sum _{k=1}^n{v}_k.$ As a new value {overscore (v)}_k is predicted, it is stored in a circular array from where it is retrieved when the corresponding v_k measurement is available. The differences between the values are used for updating the value of the variance σ². The confidence coefficient is determined as a function of |(v_k—v_k)/σ| or (v_k—v_k)²/σ² in accordance with the following table.

		Confidence
|ek/σ|	ek2/σ2	(%)
.0	.0	100.00
.1	.01	92.04
.2	.04	84.14
.3	.09	76.42
.4	.16	68.92
.5	.25	61.7
.7	.50	48.4
1.0	1.0	31.74
1.5	2.3	13.36
2.0	4.0	4.56
2.5	6.3	1.24
3.0	9.0	0.26

The confidence column is determined by the modified ERF function,

$1-\frac{1}{\sqrt{2\pi }}{\int }_{\frac{-\varepsilon }{\sigma }}^{\frac{+\varepsilon }{\sigma }}{\mathrm{\_e}}^{-\frac{t^2}{\underset{\_}{2}}}dt.$

Value ε is the absolute value of the difference between the last measured value and the corresponding predicted value. The relation,1/(1+0.77439865x+0.80155580x²+0.54110010x³+0.030646485x⁴),where x is |e_k/σ|, provides an approximation of the ERF function with errors between −0.00067 and +0.001315 for 0≦x≦3.4.

Predictions

The ATM determines the instantaneous train consist acceleration. Predicted values such as acceleration, speed, and distance are integrated starting with the present position, speed, and acceleration. If all the coefficients are already determined, the equation (58) is written asa(t)=Φ(t).  (80)

The beginning of the prediction is considered as t=0. The numerical integration is performed in steps. Thus, step i=0 corresponds to t=0, step i=1 corresponds to t=Δt, and so on. The size of Δt is capable of being equal to one second, a half second, 0.1 seconds, or smaller and larger values. A better prediction will be provided using a smaller value so that the value a₀ equals the acceleration value measured at t=0.

When determining the acceleration, the Φ(t) function uses two determined values, the current locomotive position and the speed. To maintain consistency with other relations, the function parameter used is i. The notation Φ(i) represents the value of the function determining the acceleration, using a speed value v_i and a distance d_i.

In one embodiment, the speed is determined by utilizing the known Euler method, giving the equationv_i=v_i-1+Δta_i-1.  (81)

The distance is determined by the speed value at the beginning and the end of the time interval, according tod_i=d_i-1+Δt(v_i+v_i-1)/2.  (82)

The acceleration is provided by the relation (58) symbolized as the Φ(i) function. Φ(i) uses the new position d_i to identify the length of the consist affected by elevation change and track curvature and the most recent determined v_i speed, accordinglya_i=Φ(i).  (38)

The acceleration is considered to be constant for the length Δt. The error in determining the distance d_n is approximately Δt²Φ(τ) d_n/12 where 0<τ<nΔt. When Φ(t) has negative values (braking) a shorter distance is given. Therefore, using the Euler method for approximating the stopping distance, provides smaller values than actual stopping distance.

The relative error of distance determined with the Euler method is proportional with Δt², and requires one evaluation of function Φ(i) per iteration.

In another embodiment, the speed is determined by using the known Milne method for second degree equations y″=f(x, y, y′). Each integrated parameter is determined twice. A predicted value is determined first and a corrected value is determined second. Thus, the speed prediction is given by,v_i=v_i-4+4Δt(2a_i-3−a_i-2+2a_i-1)/3,d_i=d_i-3+3(d_i-1−d_i-2)+Δt²(a_i-1−a_i-2), anda_i=Φ(i).  (84)

The correction is given by,v_i=v_i-2+Δt(a_i-2+4a_i-1+a_i)/3,d_i=d_i-1+Δt(v_i+v_i-1)/2−Δt²(a_i−a_i-1)/12, anda_i=Φ(i).  (85)

The Milne method requires two determinations of Φ(i) function per iteration and the relative error is proportional with Δt⁵.

Initial values, a₀, v₀, d₀, are the last measured values. Additionally, Milne's integration method requires a₋₁, v₋₁, d₋₁, a₋₂, v₋₂, d₋₂, a₋₃, v₋₃, d₋₃, which are the values measured in the last three steps before the integration starts. Since the data collection time interval is one second, Δt is one second.

The speed after one minute is determined by using the integration procedure presented in (83). During the integration, the function Φ(i) is evaluated keeping constant the pressure in the brake pipe and the traction force. Each step of integration determines a_i, v_i, d_i from previously determined four triplets. The integration continues for sixty steps giving the predicted value of v₆₀.

The shortest breaking distance is determined by using the same integration method until the speed v_n becomes zero. During the integration the pressure in brake pipe 32, (shown in FIG. 2), is set to a constant value corresponding to full service brake application and the traction force is set to zero. The integration continues until v_n≦0, thereby giving the predicted braking distance having a value of d_n.

Model Management

Using information from a consist description, default or initial values for model coefficients are determined. The biggest disadvantage of this method is that the errors caused by incorrect or incomplete consist information, do not decrease while new data is collected, so the quality of predictions does not improve with time.

The ATM uses the system of linear equations given above. The system of equations is updated every second with new data about the consist performance. Therefore by solving the system of equations an updated model is given every second.

The normal least squared matrix (LSM) generated for a consist is suitable to be transported and used as initial data on another consist with identical or close characteristics. Information about differences in consist structures is used for reshaping the initial normal matrix.

FIG. 3 is a chart, based on an exemplary embodiment, depicting the difference between actual stopping distance and the predicted stopping distance utilizing the ATM with no initial values. The exemplary embodiment relates to a consist having two locomotives, 50 loaded railcars and 51 empty railcars, the stopping distances are determined by using no initial information. The most important event regarding train movement and data collection are given in the following table. All determinations refer to the same stopping operation. The X axis presents the time or the “age” of the model. The Y axis presents the predicted stopping distance compared with the actual stopping distance. Each prediction is made using the ATM coefficients resulting from data collected prior to the moment of prediction.

	Time	Trip
No	(sec)	(feet)	Event	Comments
1.	6	0	Start on notch 1	Uphill orientation; train is too
				heavy and moves backward
2.	36	−62	Notch set to 2	Train keeps moving backward
3.	66	−128	Notch set to 3	Train starts moving forward
				after backing up 148 feet.
4.	126	−123	7 psi brake pipe	Train Speed is about 1 MPH
			reduction
5.	176	−76	Train stops	Train moved forward for
				71 feet of the 148 feet
				backed up.
6.	186	−76	Release brake
7.	186	−76	Notch changed from	Train starts moving forward.
			3 to 5
8.	685	8408	Notch set to 0
9.	685	8408	7 psi brake pipe	Speed is 17.56 MPH
			reduction
10.	716	9184	10 psi brake pipe
			reduction
11.	746	9808	Release brake
12.	746	9808	Notch set to 4
13.	946	13442	Notch set to 0
14.	946	13442	7 psi brake pipe
			reduction
15.	976	13862	15 psi brake pipe
			reduction
16.	1007	14076	Train stops
17.	1176	14076	Notch changed from	Brakes are still on
			0 to 2
18.	1176	14076	Brake release	The pressure starts to build in
				brake pipe.
19.	1180	14078	Train stops	Train moves 3 feet and stops
				because brakes are not
				released yet.
20.	1184	14078	Train start moving	Brakes are releasing gradually
21.	1206	14088	Notch changed from
			2 to 8
22.	2006	45382	Notch set to 0	Speed is 47.68 MPH
23.	2006	45382	Brake pipe
			reduction of
			26 psi (full service
			application)
24.	2109	49609	Train stops	Stopping distance is
				4227 feet.

In accordance with the above table, FIG. 3 shows that before the second 1400, the predicted stopping distance differs from the real value by more than ±25%. Starting with the second 1447, after traveling 20336 feet, from the previously collected data, the model predicts the stopping distance that will occur at second 2008, with an error smaller than +5%. After the second 1521, the stopping distance is predicted with an error smaller than +2%. Between time 1850 and 2000, the error of predicted stopping distance has variations between +5% and +12%. As the predicted event become closer, the error of the prediction becomes smaller than +10%. As the full service brake is applied, the error drops to −4%. Those variations in prediction accuracy are caused by the fact that the ATM determines the best approximation for the already collected data corresponding to past events, with the assumption that such data are representative for future events. When passed events do not provide enough information or are not similar with the future events, the prediction is not very accurate.

Analyzing the variation of all model coefficients determined between time 1400 and 2000, it is seen that predictions made between seconds 1520 and 1850, which are closer to 100%, are made using a set of coefficients that do not properly predict the variation of acceleration when the railcar brakes are fully applied at second 2006. Although the acceleration variation is not perfectly modeled, the estimation errors compensate for each other and the whole prediction provides a very accurate stopping distance.

FIG. 4 is a chart, for a second exemplary embodiment, showing the stopping distances determined on a consist having two locomotives, 50 loaded railcars and 51 empty railcars, using information collected in the previous embodiment. Thus, the ATM is not started from zero. The accuracy of the predicted stopping distance for the full brake application at second 2006, is shown in FIG. 4.

Since the model already has data about stopping the train with a full service brake application, the stopping distance is determined having an error between −4.2% and −2.7%.

Thus, the data collected by a train consist during one trip can be used as initial data for the ATM when the consist makes another trip with the same amount of load. For example, consists moving ore or coal perform the same trip again and again with the same consists, loaded with almost the same amount of ore in one direction and always empty in other direction.

FIG. 5 is a chart, for a third exemplary embodiment, depicting the stopping distances determined on a consist having two locomotives, 50 loaded railcars and 51 empty railcars, using information collected for the consist on a prior trip where the load was different from the present load. FIG. 5 presents the variation of the predicted stopping distance when the model is initialized with data collected by another train that is 5% heavier and its measured stopping distance is 35% shorter.

Despite those differences, the stopping distance is predicted with errors only between −4% and −10%. This range of errors is caused because although some of the initialization information was useful, it also included data that does not fit the characteristics of the consist being characterized.

FIG. 6 is a chart for a fourth exemplary embodiment, based on a consist having two locomotives, 50 loaded railcars and 51 empty railcars. FIG. 6 presents the variation of the predicted stopping distance determined with the ATM using as initialization information only some of the data collected based on values from a prior consist. More particularly, only data collected during the 300 second period that includes a full service application were used. Thus, the model makes very good predictions of the stopping distance after collecting data about the new consist for less than 10 minutes.

When the brakes are applied at second 685, the new information associated with this event helps to improve the quality of prediction of the stopping that is about 1400 seconds away at the time.

Looking at the variation of the prediction after second 2000, when the full service brake is actually applied, the quality of prediction first goes down from 100% to 96%, then, as the model acquired new data about train reaction to a full service brake, the quality moved up to 99%. Such variation is characteristic to models that improve the quality of coefficients. Since the previous prediction was 100%, it means that the global prediction was good, while the acceleration variation during the stopping was less accurate.

FIG. 7 is a chart, for a fifth exemplary embodiment, based on a consist having two locomotives, 50 loaded railcars and 51 empty railcars. FIG. 7 presents the prediction of the same stopping distance using modified initial data. Before being used for initialization, the data are adjusted according to the difference of weight between the two consists. With this new initial data, the prediction is much closer to the real stopping distance than in previous determinations. At the end of the period, as the model acquires new data, the prediction line approaches the 100% level.

In FIGS. 3 through 7, the predictions presented are of the stopping distance associated with full brake application that happens at the end of the 2006 second period. However, the ATM is capable of providing the stopping distance as the brakes are applied.

FIG. 8 is a chart, for a sixth exemplary embodiment, based on a consist having two locomotives, 50 loaded railcars and 51 empty railcars. FIG. 8 presents the predictions of the stopping distance as the brakes are fully applied instantly. In other words, it is the safest distance between two consists following each other on the same track. The predictions of the safe distance are made using a model that is not initialized.

FIG. 9 is a chart, for a seventh exemplary embodiment, based on a consist having two locomotives, 50 loaded railcars and 51 empty railcars. FIG. 9 presents the variation of the predicted stopping distance determined with the ATM using initialization information based on data collected after three trips on the same track, and using the predictions made on the third trip as references. The difference between the safe distance predicted by the ATM that has no initial data and predictions made on the third pass of the train on the same track are shown. The variation of errors for the non-initialized model is between −725 and 1335 feet.

FIG. 10 is a chart, for an eighth exemplary embodiment, based on a consist having two locomotives, 50 loaded railcars and 51 empty railcars. FIG. 10 presents the error of the estimated safe distance for the model initialized with weighted information. The variation of errors is between −410 and 268 feet.

Model Database

Referring to FIG. 1, the data collected by the ATM running on a consist is stored in database 16 from where it is retrieved and used as initial data to another consist with the same or close structure. Database 16 is accessed on the keys, such as, number of locomotives, locomotive type, number of railcars, train length (feet), train weight (tons), and load distribution along the consist.

While the invention has been described in terms of various specific embodiments, those skilled in the art will recognize that the invention can be practiced with modification within the spirit and scope of the claims.

Claims

1. A method for predicting train consist reactions to specific stimuli using a system including at least one measurement sensor located on a train consist, a data base, and a computer, the train consist including at least one locomotive and at least one railcar, said method comprising the steps of: collecting sensor data as the consist is moving;determining a consist force balance utilizing the sensor data and the computer;determining a set of consist coefficients using the computer; andpredicting train consist kinetic characteristic values using the consist force balance and the set of consist coefficients including: determining an acceleration prediction;determining a speed after one minute prediction using the acceleration prediction; anddetermining a shortest braking distance prediction using the acceleration prediction.

2. A method in accordance with claim 1 wherein said step of collecting sensor data comprises the steps of: monitoring a force applied to the consist utilizing the at least one measurement sensor;generating force data with respect to the force applied; andcommunicating the force data to the computer.

3. A method in accordance with claim 1 wherein said step of determining a consist force balance comprises the step of determining a set of consist kinetic elements.

4. A method in accordance with claim 3 wherein said step of determining a set of consist kinetic elements comprises the step of determining rolling forces according to the equation: F(rf)=M(Kr+Krvv(t))whereinF(rf) relates to the rolling forces of the train;M is the total train mass;Kr is the corrective factor for friction of the train;Krv is the dynamic corrective factor for friction of the train; andv(t) is the speed of the train as a function of time.

5. A method in accordance with claim 3 wherein said step of determining a set of consist kinetic elements further comprises the step of determining aerodynamic forces according to the equation: F(af)=Kav(t)2 whereinF(af) relates to the aerodynamic forces of the train;Ka is the corrective factor for the effect of the aerodynamic friction; andv(t) is the speed of the train as a function of time.

6. A method in accordance with claim 3 wherein said step of determining a set of consist kinetic elements further comprises the step of determining elevation caused forces according to the equation: F(ef)=M(Ke1E1(t)+Ke2E2(t)+Ke3E3(t)+Ke4E4(t))whereinF(ef) relates to the elevation caused forces of the train;M is the total train mass;Ke1 is the corrective factor for the effect of the elevation change on a first segment of the train;E1(t) is the elevation function relating to the first segment;Ke2 is the corrective factor for the effect of the elevation change on a second segment of the train;E2(t) is the elevation function relating to the second segment;Ke3 is the corrective factor for the effect of the elevation change on a third segment of the train;E3(t) is the elevation function relating to the third segment;Ke4 is the corrective factor for the effect of the elevation change on a fourth segment of the train; andE4(t) is the elevation function relating to the fourth segment.

7. A method in accordance with claim 3 wherein said step of determining a set of consist kinetic elements further comprises the step of determining braking forces caused by direction changes according to the equation: F(dbf)=M(KpCp(t)+K1C1(t))whereinF(dbf) relates to the dynamic braking force of the train;M is the total train mass;Kp is the corrective factor for the weight increase of the train;Cp(t) is the braking effect caused by the weight increase;K1 is the corrective factor for the effect of the lateral friction of the train; andC1(t) is the braking effect caused by the lateral friction.

8. A method in accordance with claim 3 wherein the at least one railcar includes at least one brake shoe, said step of determining a set of consist kinetic elements further comprises the step of determining consist brake forces caused by application of the at least one brake shoe according to the equation: F(baf)=Kb1B1(t)+Kb2B2(t)+Kb3B3(t)+Kb4B4(t)whereinF(baf) relates to the applied braking forces of the train;Kb1 is the brake function coefficient relating to a first segment of the train;B1(t) is the brake function relating to the first segment;Kb2 is the brake function coefficient relating to a second segment of the train;B2(t) is the brake function relating to the second segment;Kb3 is the brake function coefficient relating to a third segment of the train;B3(t) is the brake function relating to the third segment;Kb4 is the brake function coefficient relating to a fourth segment of the train; andB4(t) is the brake function relating to the fourth segment.

9. A method in accordance with claim 8 wherein said step of determining consist brake forces caused by application of the at least one brake shoe further comprises the steps of: determining friction coefficients of the at least one brake shoe;determining total brake application forces; anddetermining total brake release forces.

10. A method in accordance with claim 9 wherein said step of determining total brake application forces comprises the step of determining a brake application dragging force using a fast building pressure model according to the equation: Bff=min(0,max(1,(T+3.86950758*T2+0.23164628*T3)/(16367.9101+111.652789*T+27.61345048*T2−0.0026229*T3)))Bcf whereinBff is the braking force of the train for fast building pressure;T is the traction force of the train; andBcf is the brake cylinder force of the train.

11. A method in accordance with claim 9 wherein said step of determining total brake application forces comprises the step of determining a brake application dragging force using a slow building pressure model according to the equation: Bfs=min(0,max(1,(Ts+2.00986206*Ts2+0.81412194*Ts3)/(0.00067603+169.361303*Ts+8.95254599*Ts2+0.58477705*Ts3)))Bcs whereinBfs is the braking force of the train for slow building pressure;Ts is the traction force for the slow building pressure; andBcs is the brake cylinder force of the train.

12. A method in accordance with claim 9 wherein said step of determining total brake release forces comprises the step of determining brake release using a fast release model according to the equation: Rff=min(0,max(1,(t+3.86950758*t2+0.23164628*t3)/(16367.9101+111.652789*t+27.61345048*t2−0.0026229*t3)))Bcf whereinRff relates to the fast release force of the train;t is the time; andBcf is the brake cylinder force of the train.

13. A method in accordance with claim 9 wherein said step of determining total brake release forces comprises the step of determining brake release using a slow release model according to the equation: Rfs=min(0,max(1,(t+2.00986206*t2+0.81412194*t3)/(0.00067603+169.361303*t+8.95254599*t2+0.58477705*t3)))Bcs whereinRfs relates to the slow release force of the train;t is the time; andBcs is the brake cylinder force of the train.

14. A method in accordance with claim 3 wherein said step of determining a set of consist kinetic elements further comprises the step of determining dynamic brake force according to the equation: F(dbf)=KdD(t)whereinF(dbf) relates to the dynamic brake force;Kd is the corrective factor for the effect of the dynamic brake application; andD(t) is the dynamic brake force of the train.

15. A method in accordance with claim 3 wherein said step of determining a set of kinetic elements further comprises the step of determining traction force.

16. A method in accordance with claim 3 wherein said step of determining a force balance further comprises the step of summing the set of consist kinetic elements.

17. A method in accordance with claim 1 wherein said step of determining a set of consist coefficients comprises the step of using a least squares method to determine consist coefficients.

18. A method in accordance with claim 17 wherein said step of using the least squares method comprises the steps of: weighting data;solving the system; anddetermining a confidence measure.

19. A method in accordance with claim 1 wherein said step of determining an acceleration prediction comprises the steps of: determining initial values; andstoring the initial values in the database.

20. A method in accordance with claim 19 wherein said step of determining an acceleration prediction further comprises the step of determining the acceleration prediction value using a Euler method and the determined initial values.

21. A method in accordance with claim 19 wherein said step of determining an acceleration prediction further comprises the step of determining the acceleration prediction value using a Milne method and the determined initial values.

22. A system for predicting reactions of a train consist to specific stimuli, said system comprising at least one measurement sensor located on the train consist, a data base, and a computer, the train consist comprising at least one locomotive and at least one railcar, said system configured to: collect sensor data as the consist is moving;determine a consist force balance utilizing the sensor data and said computer;determine a set of consist coefficients using said computer; andpredict train consist kinetic characteristic values using the consist force balance and the set of consist coefficients wherein the system is configured to: determine an acceleration prediction;determine a speed after one minute prediction using said acceleration prediction; anddetermine a shortest braking distance prediction using said acceleration prediction.

23. A system in accordance with claim 22 wherein to collect sensor data said system further configured to: monitor a force applied to the consist utilizing said at least one measurement sensor;generate force data with respect to the force applied; andcommunicate the force data to said computer.

24. A system in accordance with claim 22 wherein to determine a consist force balance, said system further configured to determine a set of consist kinetic elements.

25. A system in accordance with claim 24 wherein to determine a set of consist kinetic elements, said system further configured to determine rolling forces according to the equation: F(rf)=M(Kr+Krvv(t))whereinF(rf) relates to the rolling forces of the train;M is the total train mass;Kr is the corrective factor for friction of the train;Krv is the dynamic corrective factor for friction of the train; andv(t) is the speed of the train as a function of time.

26. A system in accordance with claim 24 wherein to determine a set of consist kinetic elements, said system further configured to determine aerodynamic forces according to the equation: F(af)=Kav(t)2 whereinF(af) relates to the aerodynamic forces of the train;Ka is the corrective factor for the effect of the aerodynamic friction; andv(t) is the speed of the train as a function of time.

27. A system in accordance with claim 24 wherein to determine a set of consist kinetic elements, said system further configured to determine elevation caused forces according to the equation: F(ef)=M(Ke1E1(t)+Ke2E2(t)+Ke3E3(t)+Ke4E4(t))whereinF(ef) relates to the elevation caused forces of the train;M is the total train mass;Ke1 is the corrective factor for the effect of the elevation change on a first segment of the train;E1(t) is the elevation function relating to the first segment;Ke2 is the corrective factor for the effect of the elevation change on a second segment of the train;E2(t) is the elevation function relating to the second segment;Ke3 is the corrective factor for the effect of the elevation change on a third segment of the train;E3(t) is the elevation function relating to the third segment;Ke4 is the corrective factor for the effect of the elevation change on a fourth segment of the train; andE4(t) is the elevation function relating to the fourth segment.

28. A system in accordance with claim 24 wherein to determine a set of consist kinetic elements, said system further configured to determine braking forces caused by direction changes according to the equation: F(dbf)=M(KpCp(t)+K1C1(t))whereinF(dbf) relates to the dynamic braking force of the train;M is the total train mass;Kp is the corrective factor for the weight increase of the train;Cp(t) is the braking effect caused by the weight increase;K1 is the corrective factor for the effect of the lateral friction of the train; andC1(t) is the braking effect caused by the lateral friction.

29. A system in accordance with claim 24 wherein said at least one railcar comprises at least one brake shoe, and to determine a set of consist kinetic elements, said system further configured to determine consist brake forces caused by application of said at least one brake shoe according to the equation: F(baf)=Kb1B1(t)+Kb2B2(t)+Kb3B3(t)+Kb4B4(t)whereinF(baf) relates to the applied braking forces of the train;Kb1 is the brake function coefficient relating to a first segment of the train;B1(t) is the brake function relating to the first segment;Kb2 is the brake function coefficient relating to a second segment of the train;B2(t) is the brake function relating to the second segment;Kb3 is the brake function coefficient relating to a third segment of the train;B3(t) is the brake function relating to the third segment;Kb4 is the brake function coefficient relating to a fourth segment of the train; andB4(t) is the brake function relating to the fourth segment.

30. A system in accordance with claim 29 wherein to determine consist brake forces caused by application of said at least one brake shoe, said system further configured to: determine friction coefficients of said at least on brake shoe;determine total brake application forces; anddetermine total brake release forces.

31. A system in accordance with claim 30 wherein to determine total brake application forces, said system further configured to determine a brake application dragging force using a fast building pressure model according to the equation: Bff=min(0,max(1,(T+3.86950758*T2+0.23164628*T3)/(16367.9101+111.652789*T+27.61345048*T2−0.0026229*T3)))Bcf whereinBff is the braking force of the train for fast building pressure;T is the traction force of the train; andBcf is the brake cylinder force of the train.

32. A system in accordance with claim 30 wherein to determine total brake application forces, said system further configured to determine a brake application dragging force using a slow building pressure model according to the equation: Bfs=min(0,max(1,(Ts+2.00986206*Ts2+0.81412194*Ts3)/(0.00067603+169.361303*Ts+8.95254599*Ts2+0.58477705*Ts3)))Bcs whereinBfs is the braking force of the train for slow building pressure;Ts is the traction force for the slow building pressure; andBcs is the brake cylinder force of the train.

33. A system in accordance with claim 30 wherein to determine total brake release forces, said system further configured to determine brake release using a fast release model according to the equation: Rff=min(0,max(1,(t+3.86950758*t2+0.23164628*t3)/(16367.9101+111.652789*t+27.61345048*t2−00026229*t3)))Bcf whereinRff relates to the fast release force of the train;t is the time; andBcf is the brake cylinder force of the train.

34. A system in accordance with claim 30 wherein to determine total brake release forces, said system further configured to determine brake release using a slow release model according to the equation: Rfs=min(0,max(1,(t+2.00986206*t2+0.81412194*t3)/(0.00067603+169.361303*t+8.95254599*t2+0.58477705*t3)))Bcs whereinRfs relates to the slow release force of the train;t is the time; andBcs is the brake cylinder force of the train.

35. A system in accordance with claim 24 wherein to determine a set of consist kinetic elements, said system further configured to determine dynamic brake force according to the equation: F(dbf)=KdD(t)whereinF(dbf) relates to the dynamic brake force;Kd is the corrective factor for the effect of the dynamic brake application; andD(t) is the dynamic brake force of the train.

36. A system in accordance with claim 24 wherein to determine a set of kinetic elements, said system further configured to determine traction force.

37. A system in accordance with claim 24 wherein to determine a force balance, said system further configured to sum said set of consist kinetic elements.

38. A system in accordance with claim 22 wherein to determine a set of consist coefficients, said system further configured to use a least squares method to determine consist coefficients.

39. A system in accordance with claim 38 wherein to use the least squares, said system further configured to: weight data;solve the system; anddetermine a confidence measure.

40. A system in accordance with claim 22 wherein to determine an acceleration prediction, said system further configured to: determine initial values; andstore the initial values in said database.

41. A system in accordance with claim 40 wherein to determine an acceleration prediction, said system further configured to determine the acceleration prediction value using a Euler method and said determined initial values.

42. A system in accordance with claim 19 wherein to determine an acceleration prediction, said system further configured to determine the acceleration prediction value using a Milne method and the determined initial values.

43. A method for determining a force balance for a train consist using a system including at least one measurement sensor located on the train consist, a data base, and a computer, the train consist including at least one locomotive and at least one railcar, the railcar including at least one brake shoe, said method comprising the steps of: monitoring a force applied to the consist utilizing the at least one measurement sensor;generating force data with respect to the force applied;communicating the force data to the computer;determining rolling forces according to the equation F(rf)=M(Kr+Krvv(t)),determining aerodynamic forces according to the equation F(af)=Kav(t)2,determining elevation caused forces according to the equation F(ef)=M(Ke1E1(t)+Ke2E2(t)+Ke3E3(t)+Ke4E4(t)),determining braking forces caused by direction changes according to the equation F(dbf)=M(KpCp(t)+K1C1(t));determining consist brake forces caused by application of the at least one brake shoe according to the equation F(baf)=Kb1B1(t)+Kb2B2(t)+Kb3B3(t)+Kb4B4(t);determining brake application dragging force using a fast building pressure model according to the equation: Bff=min(0,max(1,(T+3.86950758*T2+0.23164628*T3)/(16367.9101+111.652789*T+27.61345048*T2−0.0026229*T3)))Bcf;determining brake application dragging force using a slow building pressure model according to the equation: Bfs=min(0,max(1,(Ts+2.00986206*Ts2+0.81412194*Ts3)/(0.00067603+169.361303*Ts+8.95254599*Ts2+0.58477705*Ts3);determining brake release using a fast release model according to the equation: Rff=min(0,max(1,(t+3.86950758*t2+0.23164628*t3)/(16367.9101+111.652789*t+27.61345048*t2−0.0026229*t3)))Bcf,determining brake release using a slow release model according to the equation: Rfs=min(0,max(1,(t+2.00986206*t2+0.81412194*t3)/(0.00067603+169.361303*t+8.95254599*t2+0.58477705*t3)))Bcs determining dynamic brake force according to the equation F(dbf)=KdD(t),determining traction force; anddetermining a final solution according to the equation: F(t)=M(Kr+Krvv(t))+Kav(t)2+MKe1E1(t)+MKe2E2(t)+MKe3E3(t)+MKe4E4(t)+MKpCp(t)+MK1C1(t)+Kb1B1(t)+Kb2B2(t)+Kb3B3(t)+Kb4B4(t)+Kr1R1(t)+Kr2R2(t)+Kr3R3(t)+Kr4R4(t)+KdD(t)+KtT(t)