METHOD FOR CONTROLLING THE QUALITY OF INDUSTRIAL PROCESSES, IN PARTICULAR LASER-WELDING PROCESSES

Abstract

A system and a method for controlling the quality of an industrial process, of the type that comprises the steps of: providing one or more reference signals for the industrial process; acquiring one or more real signals that are indicative of the quality of said industrial process; and comparing said one or more reference signals with said one or more real signals in order to identify defects in said industrial process. According to the invention, the method moreover comprises the operations of: obtaining a transformed signal from said reference signal; obtaining a transformed signal from said real signal; and calculating energies of said transformed reference signal and said real signal, respectively, said comparison operation comprising; comparing with one another said energies of said transformed reference signal and said transformed real signal, respectively, in order to extract corresponding time-frequency distribution for selected frequency values; calculating energies of said time-frequency distributions; and comparing the energies of said time-frequency distributions with threshold values in order to identify energy values associated to defects.

Description

BACKGROUND OF THE INVENTION

The present invention relates to methods for controlling the quality of an industrial process, comprising the steps of:

• • providing one or more reference signals for an industrial process;
  • acquiring one or more real signals that are indicative of the quality of said industrial process; and
  • comparing said one or more reference signals with said one or more real signals in order to identify defects in said industrial process.

Monitoring of the defects in industrial processes assumes increasing economic importance on account of its impact on the analysis of quality of industrial products. The possibility of obtaining an on-line and automatic assessment of the quality of an industrial process presents many advantages both from the economic point of view and from the standpoint of the speed of the process. Desirable characteristics of the system are hence:

• • on-line and real-time processing; and
  • capacity for accurate recognition of the main defects in production.

Currently the problem of recognition of the quality of an industrial process, and consequently the identification of defects, is approached by an inspection carried out off line by skilled staff, or else using automatic methods, which, by means of sensors, identify only some of the defects listed above, in a way that is far from satisfactory and is moreover sensitive to the different settings of the machine.

There are known methods and systems for controlling the quality in industrial processes, for example applied to on-line monitoring of the laser-welding process, in particular in the case of welding of sheet metal. The control system is able to assess the presence of porosities in the welding area or else, in the case of butt-welded thin sheet metal, the presence of defects due to overlapping or poor jointing of the sheet metal.

The above used systems base quality control on a comparison between the signals detected during the process and one or more predetermined reference signals indicating a good-quality weld. Said reference signals, which usually range in number between two and ten, are arranged starting from a number of samples of good-quality welds. Obviously, said mode of procedure implies the presence of a skilled operator who is able to certify the goodness of the weld at the moment of creation of the reference signals, and involves expenditure in terms of time and sometimes also in terms of waste of material (used for making the samples necessary for obtaining the reference signals). In some cases there are also pre-arranged reference signals indicating a defective weld, this, however, involving additional problems and difficulties.

From the European patent application No. EP-A-1275464 filed in the name of the present applicant, it is known to split into blocks the signal acquired via photodiode, which collects the radiation emitted by a welding spot, calculating the mean of the signal in each block sampled and considering the blocks having a value smaller than or equal to the offset of the photodiode as indicating the presence of a defect. Said method eliminates the need for the reference signal; however, it enables only a very approximate detection of the defects.

SUMMARY OF THE INVENTION

The aim of the present invention is to overcome all the aforesaid drawbacks.

In order to achieve such aim, the object of the present invention is a method for controlling the quality of industrial processes which has the characteristics indicated at the beginning and is further characterized in that said method further comprises the operations of:

• • obtaining a transformed signal from said reference signal;
  • obtaining a transformed signal from said real signal; and
  • calculating energies of said transformed reference signal and said real signal,

said comparison operation comprising:

• • comparing with one another said energies of said transformed reference signal and said transformed real signal for extracting corresponding time-frequency distributions for selected frequency values;
  • calculating energies of said time-frequency distributions; and
  • comparing the energies of said time-frequency distributions with threshold values in order to identify the energy values associated to defects.

In the preferred embodiment, said steps of obtaining a transformed signal from said reference signal and of obtaining a transformed signal from said real signal comprise a filtering operation by means of the application of a discrete wavelet transform (DWT), whilst said operation of comparing said energies of said transformed reference signal and said transformed real signal for obtaining corresponding time-frequency distributions comprises performing a calculation of the conjugate of the Fourier transform of the envelope of the real signal and of the envelope of the normalized signal, to obtain a real conjugated transformed signal and a reference conjugated transformed signal, respectively, as well as comparing the energies of the reference signal and of the real signal, extracting the frequency values for which the energy of the real signal is greater than that of the reference signal.

Of course, a further object of the invention is the system for controlling the quality of industrial processes that implements the method described above, as well as the corresponding computer product, directly loadable into the memory of a computer, such as a processor, and comprises software code portions for performing the method according to the invention when the product is run on a computer.

BRIEF DESCRIPTION OF THE DRAWINGS

Further characteristics and advantages of the invention will emerge from the ensuing description, with reference to the annexed drawings, which are provided purely by way of non-limiting example and in which:

FIG. 1 is a block diagram representing a system implementing the method according to the invention;

FIG. 2 shows a detail of the system of FIG. 1;

FIGS. 3, 4 and 5 are flowcharts representing operations of the method according to the invention; and

FIG. 6 is a diagram of quantities processed by the method according to the invention.

DETAILED DESCRIPTION OF THE INVENTION

The method according to the invention will now be exemplified with reference to a laser-welding method. Said laser-welding method constitutes merely, however, just one non-limiting example of an industrial process to which the method for controlling the quality of industrial processes according to the invention can be applied.

With reference to FIG. 1, the reference number 1 indicates, as a whole, a system for controlling the quality of a laser-welding process. The example relates to the case of two pieces of sheet metal 2, 3, which are welded by means of a laser beam. The number 4 designates as a whole the focussing head, including a lens 5, at which there arrives the laser beam originated by a laser generator (not illustrated) and reflected by a half-reflecting mirror 6, after passage through a lens L. The radiation E emitted by the welding area passes through the half-reflecting mirror 6 and is detected by a sensor 7 constituted by a photodiode that is able to send its outgoing signal to an electronic control and processing unit 8, associated to a personal computer 9.

In a concrete embodiment, the half-reflecting mirror 6 used is a mirror made of ZnSe, with a diameter of 2 ins and a thickness of 5 mm. The sensor 7 is made up of a photodiode with a spectral response of between 190 nm and 1100 nm and an active area of 1.1×1.1 mm and a quartz window.

FIG. 2 illustrates in greater detail the electronic control and processing unit 8 associated to the personal computer 9. Said processing unit 8 comprises an anti-aliasing filter 11, which operates on the signal sent by the sensor 7. There is then envisaged an acquisition card 12 equipped with an analog-to-digital converter, which samples the filtered signal and converts it numerically. Said acquisition card 12 is preferably directly associated to the personal computer 9.

Once again in the case of a concrete embodiment, the acquisition card 12 is a data-acquisition card of the type PC card NI 6110E, with a maximum frequency of acquisition of 5 Msamples/s.

The anti-aliasing filter 11 carries out a filtering of the signal by means of a low-pass filter (for example a Butterworth IIR filter).

In the personal computer 9, according to the invention there is implemented a method for quality control, which is based upon a comparison of a real signal X_real, acquired via the photodiode 7 and a reference signal X_ref, representing a defective weld, stored in said personal computer 9.

The reference signal, designated by X_ref(t) is acquired at a frequency of acquisition f_s, and hence, according to Nyquist's theorem, has associated to it a frequency band of the signal having the value of f_s/2, whilst the number of samples acquired for the reference signal X_ref(t) is N.

FIG. 3 illustrates a flowchart representing the operations performed on the reference signal X_ref(t).

In a first step 100, an operation of filtering of the reference signal X_ref(t) is performed by means of the application of a discrete wavelet transform (DWT). At output from the step 100 there is thus obtained a signal X_ref—DWT having N/2 samples in the band 0:f_s/4.

Subsequently, a Hilbert-transform operation is applied to the signal X_ref—DWT in a step 101, to obtain a complex analytical signal X_ref—HIL, having N/2 samples and null negative frequencies.

A normalization operation is applied to said analytical signal X_ref—HIL, in a step 102, which produces at output a normalized signal X_ref—norm.

On said normalized signal X_ref—norm there is then performed, in a step 103, an operation of calculation of an envelope of the normalized signal, designated by X_{ref—inv—norm}, whilst in a step 104, a fast-Fourier-transform (FFT) operation is applied to said envelope of the normalized signal X_{ref—inv—norm}, to obtain a transformed envelope X_{ref—inv—norm}.

Finally, in a step 105, an operation of calculation of the energy of the reference signal, designated by E_ref, is performed by applying the relation:

∫|X_{ref—inv—norm}(t)|²dt=∫|X_{ref—inv—norm}(f)|²df   (1)

As regards the real signal X_real(t), also this is acquired at a frequency of acquisition f_s, and thus, according to Nyquist's theorem, has associated to it a frequency band of the signal having a value of f_s/2, whilst the number of samples acquired for the real signal X_real(t) is N.

FIG. 4 illustrates a flowchart representing the operations performed on the real signal X_real(t).

In particular, represented in FIG. 4 is a first step 200, in which an operation of filtering of the real signal X_real(t) is performed by the application of a DWT. At output from step 200, there is thus obtained a signal X_real—DWT having N/2 samples in the band 0:f_s/4.

A fast-Fourier-transform operation is performed on said signal X_real—DWT, in a step 211, to obtain a transformed signal FFT_—real, which, subsequently, in a step 212, is normalized, to obtain a transformed normalized signal FFT_{—real—norm}.

In a step 250, an operation of calculation of a mean frequency f₀ is performed on the transformed normalized signal FFT_{—real—norm}, according to the relation:

f ₀ =∫f*FFT_{—real—norm}(f)*FFT_{—real—norm}(f)df   (2)

In a step 251, an operation of calculation of a standard deviation B is performed, according to the relation:

B=(∫f²*FFT_{—real—norm}*FFT_{—real—norm}df−f₀²)^1/2   (3)

In a step 252, there are then calculated a lower band F_Sn==(f₀−B/2) and an upper band F_Dx=(f₀+B/2).

In parallel, in a step 201, a Hilbert-transform operation is applied to the signal X_real—DWT, to obtain a complex analytical signal X_real—HIL, which has N/2 samples and having null negative frequencies.

In a step 202, a normalization operation is applied to said analytical signal X_real—HIL, which produces at output a normalized signal X_real—norm.

On said normalized signal X_real—norm there is then performed, in a step 203, an operation of calculation of the envelope, designated by X_{real—inv—norm}, whilst, in a step 204, a fast-Fourier-transform operation (FFT) is applied to said envelope of the normalized signal X_{real—inv—norm}, to obtain a transformed envelope X_{real—inv—norm}.

Finally, in a step 205, an operation of calculation of an energy of the real signal E_real is performed by applying the following relation:

∫|x_{real—inv—norm}(t)|²dt=∫|X_{real—inv—norm}(f)|²df   (4)

The operations of calculation of the energies E_real and E_ref are performed in a band delimited between the lower band F_Sn and the upper band F_Dx calculated in step 252. In greater detail, the calculation is performed on the band thus delimited, considering frequency steps, for example of one hertz, i.e.,

∫_F^s|X_{real—inv—norm}(f)|²df ∫_step^F—DX|X_{real—inv—norm}(f)|²df

∫_F^s|X_{real—inv—norm}(f)|²df ∫_step^F—DX|X_{ref—inv—norm}(f)|²df

In this way, the operation of calculation of the energies E_ref and E_real produces two respective vectors, namely a vector of energies of the reference signal Energy_Ref_step (1, . . . k), and a vector of energies of the real signal Energy_Real_step (1, . . . k), both comprising k frequency values.

Subsequently, a procedure of calculation of the quadratic time-frequency distributions is performed, illustrated in the flowchart of FIG. 5, which comprises the following operations:

• • in a step designated by 300, calculating the conjugates of the fast-Fourier transforms (FFTs) of the envelope of the real signal X_{real—inv—norm}(f) and of the envelope of the reference signal X_{ref—inv—norm}(f), to obtain conjugated transformed signals, namely, the real conjugated transformed signal X*_{real—inv—norm}(f) and the reference conjugated transformed signal X*_{ref—inv—norm}(f), respectively;
  • in a step 301, considering the energies of the reference signal E_ref and of the real signal E_real, represented, respectively, by the vector of energies of the reference signal Energy_Ref_step (1, . . . k) and the vector of energies of the real signal Energy_Real_step (1, . . . k), and, for each element k of said two vectors, assessing whether the following criterion is satisfied:

Energy_Real_step (1, . . . k)>Energy_Ref_step (1 . . . k)   (5)

This operation can be appreciated also with reference to the graph of FIG. 6, which shows the amplitudes of the energy of the reference signal E_ref and of the energy of the real signal E_real (thicker line) as a function of the frequency;

• • if the criterion (5) is satisfied, in a step 302 carrying out an operation of extraction of the frequency value for which said criterion (5) is verified, said value being indicated as f_e; according to how many times the condition is satisfied, obtaining up to a maximum of k frequency values f_e; FIG. 6 shows the regions corresponding to the frequency values f_e for which the criterion (5) is satisfied;
  • in a step 303, constructing a matrix M, the rows of which are represented by the extracted frequency values f_e, whilst the columns are represented by N/2 temporal values t₁ . . . t_N/2 of the signal at output from the DWT operation 200;
  • in a step 304, for each row of the matrix M, calculating a quadratic time-frequency distribution both for the reference signal, designated by Tfd_ref, and for the real signal, designated by Tfd_real, using the Margenau-Hill relation, namely

Tfd _real=Real(x_real—DWT(t)·X_{real—inv—norm}*(f)·and^−j2πf)   (6)

Tfd _ref=Real(x_ref—DWT(t)·X_{ref—inv—norm}*(f)·and^−j2πf)   (7)

• • then, in a step 305, calculating, for both the reference signal and the real signal, energies associated to the distributions for each instant of time, designated as Et_ref and Et_real, respectively; and
  • then, in a step 306, calculating a maximum value of the energy max_Tfd_ref for the time-frequency distribution of the reference Tfd_ref.

Finally, in order to obtain an estimate of the defects, in a step 307 each temporal value of the energy Et_real of the quadratic time-frequency distribution of the real signal Tfd_real is compared with the maximum value of the energy max_Tfd_ref.

If said value of energy of the quadratic time-frequency distribution of the real signal Tfd_real exceeds the maximum value of the energy max_Tfd_ref, this means that there is a defect on that time co-ordinate.

In this way, it is possible to locate the defects in time.

Of course, without prejudice to the principle of the invention, the details of construction and the embodiments may vary widely with respect to what is described and illustrated herein purely by way of example, without thereby departing from the scope of the present invention.

Claims

1. A system for controlling quality of a manufacturing process for recognizing defects in result of a production, comprising: sensor means for detecting one or more process parameters; andan electronic control and processing unit for processing the signals issued by said sensor means,wherein said electronic control and processing unit for processing the signals issued by said sensor means implements a method for controlling the quality of the manufacturing process comprising: acquiring, by detecting one or more quantity of the production of said manufacturing process, one or more real signals produced by a machine involved in the manufacturing process, where the one or more real signals are indicative of the quality of said manufacturing process;providing one or more reference signals corresponding to said one or more real signals and indicative of a good-quality process for the manufacturing process;obtaining a transformed signal from said reference signal;obtaining a transformed signal from said real signal;converting each of the transformed real signal and the transformed reference signal into a respective vector of energies, where the vector of energies comprises an integer number k of frequency values; andcomparing said one or more reference signals with said one or more real signals in order to identify the defects in the result of the production,wherein said comparing comprises: comparing said vector of energies of said transformed reference signal with said vector of energies of said transformed real signal and selecting frequency values among said k frequency values on the basis of said comparison, respectively, for extracting corresponding time-frequency distributions for said selected frequency values;calculating energies of said time-frequency distributions; andcomparing the energies of said time-frequency distributions with threshold values in order identify energy values associated with said defects, and the corresponding time coordinates of said defects.

2. The system according to claim 1, wherein said industrial process is a laser-welding process.

3. A computer-program product directly loadable into the memory of a computer and comprising software code portions for performing a method of controlling quality of a manufacturing process for recognizing defects in result of a production, when the product is run on a computer, wherein the method comprises: acquiring, by detecting one or more quantity of the production of said manufacturing process, one or more real signals produced by a machine involved in the manufacturing process, where the one or more real signals are indicative of the quality of said manufacturing process;providing one or more reference signals corresponding to said one or more real signals and indicative of a good-quality process for the manufacturing process;obtaining a transformed signal from said reference signal;obtaining a transformed signal from said real signal;converting each of the transformed real signal and the transformed reference signal into a respective vector of energies, where the vector of energies comprises an integer number k of frequency values; andcomparing said one or more reference signals with said one or more real signals in order to identify the defects in the result of the production,wherein said comparing comprises: comparing said vector of energies of said transformed reference signal with said vector of energies of said transformed real signal and selecting frequency values among said k frequency values on the basis of said comparison, respectively, for extracting corresponding time-frequency distributions for said selected frequency values;calculating energies of said time-frequency distributions; andcomparing the energies of said time-frequency distributions with threshold values in order identify energy values associated with said defects, and the corresponding time coordinates of said defects.