Patent Application
20050187700
The present invention relates to systems and methods for diagnosing internal combustion engines and, more particularly, to systems and methods for diagnosing and calibrating internal combustion engines using a variety of engine sensors.
Recent legislative requirements imposed by the Environmental Protection Agency demand the ability to conduct on-line diagnosis of internal combustion engine performance to ensure compliance with exhaust gas emissions regulations. One such variable that provides an excellent indication of engine performance is the indicated torque generated by each cylinder during the course of the combustion process. There are a number of approaches that may be used to calculate torque, most of which rely on a combination of knowledge from a variety of engine sensors. Also, torque calculations are so complex that several simultaneous measurements are often utilized to ensure accurate and reliable calculations. For example, one approach relies on fuel injector control settings and sensors to indicate the engine's torque level. If one injector fails, the prediction may lose considerable accuracy. The problem may go undetected except perhaps by an operator who recognizes the power loss, unless there is sensor information indicating actual injector performance. Unfortunately, production-intent injector instrumentation is too costly, so an implicit injector performance measure currently is the most viable practical option.
Instead of relying on fuel injector control settings, torque may be calculated based on the output of camshaft and crankshaft speed sensors. Since most modern internal combustion engines include a redundancy of camshaft and crankshaft speed sensors, these torque calculations are typically easier to compute and more reliable. If one sensor fails, its failure is detected and a backup sensor is used.
Recently, engine manufacturers have began to compute torque as a function of cylinder pressure. In this approach, cylinder pressure during combustion is used to compute an instantaneous crankshaft speed which is then converted to torque. The ratio of two cylinder pressure measurements (e.g., one at top dead center (TDC) and one at 60° before TDC) may also be used to compute torque. The measured pressure ratio in one or more cylinders is compared to an optimal pressure ratio for the specific engine operating conditions, and one or more injectors may be trimmed (i.e., the air-fuel ratio is modified) to optimize engine operation. The process of achieving target torque by evaluating pressure ratios has been found to be less complicated than the previously discussed methods because fewer calculations must be performed and failed sensors are more readily identified. Hardware or virtual in-cylinder pressure sensing also provides other measures not available from rotational crankshaft speed. For example, in-cylinder pressure sensing may be used to identify misfiring circuits and calculate combustion noise. Cylinder pressure may also be used to calculate and optimize the mass of air present in a cylinder, and air density in a cylinder.
Given the many methods for calculating torque, and the complexity of the calculations, engine manufacturers are constantly looking for new ways to improve the accuracy of the calculations. Lately, neural networks have been used to further improve accuracy of prior art torque estimating systems. For example, U.S. Pat. No. 6,234,010 to Zavarehi et al. discloses a method for detecting torque of a reciprocating internal combustion engine with the use of a neural network including the steps of: sensing rotational crankshaft speed for a plurality of designated crankshaft rotational positions over a predetermined number of cycles of rotation for each crankshaft position; determining an average crankshaft speed fluctuation for each crankshaft position; determining information representative of crankshaft kinetic energy variations due to each firing event and each compression event in the cylinder; determining information representative of crankshaft torque as a function of the crankshaft kinetic energy variations and the average crankshaft speed; and outputting a representative crankshaft torque signal from a neural network. Since the system disclosed in this reference computes kinetic energy variations due to combustion and compression events, two inputs for each cylinder and an input for average crankshaft speed must be entered into the neural network. This results in a very complicated, processor-intensive network calculation.
What is desirable is an accurate system and method capable of determining torque, cylinder misfires, and other engine operations that rely on a small number of engine operation measurements and do not require an excessive processing capability.
A method for determining a predetermined operating condition of an internal combustion engine is disclosed. In operation, the method measures a cylinder pressure in at least one combustion chamber at a predetermined point in a combustion cycle. Next, the method determines at least a first value for an operating parameter of the engine using the measured cylinder pressure, determines a second value for the operating parameter of the engine using data received from at least one engine sensor, and then generates a predetermined signal if a difference between the first value and the second value has a predetermined relationship. An apparatus and a machine-readable medium are also provided to implement the disclosed method.
For the purposes of promoting an understanding of the principles of the invention, reference will now be made to the embodiments illustrated in the drawings and specific language will be used to describe the same. It will nevertheless be understood that no limitation of the scope of the invention is thereby intended. The invention includes any alterations and further modifications in the illustrated devices and described methods and further applications of the principles of the invention that would normally occur to one skilled in the art to which the invention relates.
Referring now to
Referring now to
The control routine according to one exemplary embodiment of the present invention for measuring torque, misfires, and/or other operations of an internal combustion engine is shown in
After storing PT, control transfers to block 308, where the crank angle sensor 2 again measures the crank angle of the cylinder crankshaft and generates an output signal S1 to ECM 10 indicating the measured crank angle. In block 310, a query is made to determine if the crank angle is at a second predetermined angle, such as, 25° after bottom dead center (ABDC). Once it is determined that the crank angle is 25° ABDC, control is transferred to block 312 to store the cylinder pressure PB of the next cylinder (e.g., cylinder #2) (indicated by the signal S2) as measured by cylinder pressure sensor 4 in the memory 14.
Discrete pressure samples taken during the compression stroke may be used to determine the mass of air present in the cylinder. If this mass is determined to be outside of a desired range, intake or exhaust valve actuation or turbocharger operation may be at fault. If necessary, appropriate modification to the engine performance may be made. For example, the intake valve, exhaust valve and/or turbocharger may be calibrated (or trimmed) to yield the target value.
Discrete pressure samples taken during the power stroke may be used to calculate heat release in the cylinder to provide information about the fuel injection event. If the heat release is excessive or too low, for example, the timing and duration of injection pulses may be trimmed to yield a desired value.
In engines in which stroke overlap may be controlled (variable valve timing), discrete pressure samples taken during the overlap period of intake and exhaust valve opening may be used to calculate the amount of residual gas to be used in emissions/performance prediction algorithms. If the sampled pressure amount is outside of a predetermined range, for example, intake or exhaust valve actuation or turbocharger operation may be calibrated or trimmed.
In addition to relying on discrete pressure samples, the above calculations may be based upon sensor inputs. For example, a volumetric efficiency (VE) table may have axes for engine rpm (deduced, for example, from a timing sensor) and air density for fixed valve events. The VE table may have additional axes for flexible valve events. Air density is dependent on intake manifold temperature (sensor) and pressure (sensor) readings. The rule for target air mass may be that it fall within a predetermined range (e.g., +/−5%) of the value deduced via the VE table. Likewise, fuel and coolant temperatures may additionally be required to find the expected ignition delay from a lookup table. Ignition delay may be required to calculate whether or not injection timing and duration match target values in another lookup table (engine rpm, mass air, ambient conditions, and mass fuel are likely axes). In many cases, the sensor input can be from either a virtual or hardware sensor. The target may be two-fold: first trim every cylinder to perform the same, and second, trim the array of cylinders to match the target from the lookup table.
When the engine is operating at low speed and light loads, a number of factors combine to produce speed patterns that appear chaotic. Among these factors are gear lash, engine governor settings, and false gear tooth detection. One exemplary embodiment of the present invention uses a radial basis neural network (RBNN) to model known speed patterns at various levels of individual cylinder power and then uses pattern recognition to more accurately characterize engine performance during periods of seemingly random engine behavior. An RBNN is a neural network model based preferably, on radial basis function approximators, the output of which is a real-valued number representing the estimated engine torque at a designated test point. When using an RBNN, cylinder pressure data is compressed into integrated measures, as use of discrete samples would require an excessive number of model inputs. A second exemplary embodiment may use a back propagation or other neural network. Referring to
Since the illustrative, but non-limiting, internal combustion engine 12 has four cylinders, and torque magnitude is determined as a function of cylinder pressure variation due to combustion and compression effects and average crankshaft speed, the RBNN for engine torque may at least include 4 (the number of cylinders) times X (pressure variation can be described by X number of variables) inputs, plus inputs for injection timing, IMT, etc. The cells in the input layer normalize the input signals received (preferably, between −1 and +1) and pass the normalized inputs to Gaussian processing cells in the hidden layer. When the weight and threshold factors have been set to correct levels, a complex stimulus pattern at the input layer successively propagates between hidden layers, to result in a simpler output pattern. The network is “taught” by feeding it a succession of input patterns and corresponding expected output patterns. The network “learns” by measuring the difference (at each output unit) between the expected output pattern and the pattern that it just produced. Having done this, the internal weights and thresholds are modified by a learning algorithm to provide an output pattern which more closely approximates the expected output pattern, while minimizing the error over the spectrum of input patterns. Network learning is an iterative process, involving multiple “lessons”. Neural networks have the ability to process information in the presence of noisy or incomplete data and still generalize to the correct solution.
As an alternative method, using a fixed-point processor, a linear neural network approach can be used. In the linear neural network approach, the inputs and outputs are in binary −1 (or 0)+1 format, rather than the real-valued input and output data used in the radial basis neural network. With this approach, torque magnitude is determined to be the highest-valued output.
In a second exemplary embodiment of the present invention, RBNN 400 may be used to identify combustion noise (knocks). As is known in the art, the knock signal is typically generated when the cylinder pressure approaches the maximum value. While the frequency range of the knock signal varies with the inner diameter of the cylinder, it generally exceeds 5 kHz. Therefore, by passing the cylinder pressure waveform generated by RBNN 400 through a high-pass filter whose cutoff frequency is around 5 kHz, it becomes possible to extract only the knock signal. Since combustion knock also tends to indicate intense combustion temperatures that promote production of various Nitrogen Oxides (NOx), RBNN 400 may also be used to control NOx production.
Industrial Applicability
While engine 15 is designed to achieve substantially the same combustion event in each cylinder for a given set of engine conditions, in actuality, the combustion event within each cylinder will vary from cylinder to cylinder due to manufacturing tolerances and deterioration-induced structural and functional differences between components associated with the cylinders. Therefore, by monitoring the variability in the pressure ratio in the individual cylinders, the engine control system 16 can separately adjust the air-fuel ratio within the different cylinders to balance the performance of the individual cylinders. Similarly, by comparing the pressure of the individual cylinders and their variations to predetermined target pressures, the engine control system 16 of the present invention can accurately compute torque and other measurements, while also detecting poorly functioning or deteriorating components.
The present invention may be advantageously applicable in performing diagnostics and injector trim using in-cylinder pressure sensing. With the implementation of complex injection and air systems on internal combustion engines comes the difficulty of calibration and diagnostics. Some calibration can take place at the component level at each element's time of manufacture (component calibration). Other calibrations need to take place once the components have been assembled into the system (system calibration). System calibration can sometimes eliminate the need for component calibrations, thus saving the time/expense of redundant operations. This method includes the advantage of providing the capability to perform on-line diagnostics and system calibration using in-cylinder pressure sensing.
Another aspect of the described system may be the advantage of eliminating external measuring devices such as dynamometers. The representative crankshaft torque can be responsively produced and communicated to a user, stored and/or transmitted to a base station for subsequent action. This present invention can be utilized on virtually any type and size of internal combustion engine.
Yet another aspect of the described invention may be the benefit provided through the use of a neural network to model torque, combustion knocks and misfires. The use of neural networks permits the present invention to provide accurate and prompt feedback to a control module and/or system users.
Benefits of the described system are warranty reduction and emissions compliance. More accurate monitoring of the engine system will allow narrower development margins for emissions, directly resulting in better fuel economy for the end user.
While the invention has been illustrated and described in detail in the drawings and foregoing description, the same is to be considered as illustrative and not restrictive in character. It should be understood that only exemplary embodiments have been shown and described and that all changes and modifications that come within the spirit of the invention are desired to be protected.
| Number | Date | Country | |
|---|---|---|---|
| Parent | 10145103 | May 2002 | US |
| Child | 11108650 | Apr 2005 | US |
