Quality Zhaodong analysis system

Through the quality cause analysis system, using the interpretable artificial intelligence and Taguchi method, the machine and product data can be monitored in real time, the contribution of product specifications and the machine parameters are analyzed, which solves the problem of real-time monitoring and analysis of quality defects in the existing technology. The problem has improved production efficiency and quality control effect.

CN118917706APending Publication Date: 2024-11-08罗明琇 +2
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410539809.5
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Priority Date
2023-05-02
Filing Date
2024-04-30
Publication Date
2024-11-08

AI Technical Summary

Technical Problem

Existing quality inspections are difficult to monitor product quality in real time during production, and it is difficult to analyze the causes of quality defects, resulting in reduced production efficiency, increased costs and unstable product quality.

Method used

A quality cause analysis system was designed to combine the optimal machine evaluation and Taguchi method through interpretable artificial intelligence methods to monitor the machine status and product measurement data in real time, establish a quality prediction model, analyze the contribution degree of product specifications, and Optimize machine parameters to improve production efficiency.

Benefits of technology

It realizes real-time monitoring of product quality during the production process, early warning and adjustment of machine parameters, reduces the defective product rate, improves production efficiency and quality control effect, and reduces the time and cost required for process optimization.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118917706A_ABST
    Figure CN118917706A_ABST
Patent Text Reader

Abstract

The invention discloses a mass zhaodontology analysis system. The mass zhaodontology analysis system comprises an input device, a storage device, a processing device and an output device, the storage device stores machine state data, product measurement data and a plurality of algorithms. The processing device accesses the storage device to establish a Zhaodondue analysis model, and the Zhaodondue analysis model comprises a quality prediction module, a network interpretation module and an optimal machine prediction module. The quality prediction module predicts whether the product measurement data accords with quality detection regulations. And the network interpretation module measures the contribution degree value of the state variable through an interpretable artificial intelligence algorithm. The optimal machine prediction module calculates an optimal machine evaluation value, and selects a plurality of pieces of machine state data closest to the specification center value as optimal machine data.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to a quality cause analysis system, and more particularly to a quality cause analysis system that combines an explainable artificial intelligence method with an optimal machine evaluation and a Taguchi method to obtain the best contribution degree and optimal parameters of a machine. Background Art

[0002] The global manufacturing industry is rapidly developing towards big data and intelligence, and the intelligent transformation of the manufacturing industry has a great impact on the competitiveness of the industry. The current manufacturing industry can use automated devices to carry out production and manufacturing, effectively improving productivity, but it is impossible to grasp the quality status in real time, and it is even more impossible to reduce the labor cost and time consumed by quality inspection. After the entire batch of products is completed, random inspections are conducted to confirm whether there are defects or defective products. In this way, not only is it easy to generate the cost of a whole batch of defective products, but it is also likely to reduce the overall product quality level, reduce production efficiency and yield, and increase the risk of damage to the company's reputation.

[0003] In the process of intelligent transformation, the introduction of artificial intelligence technology to assist enterprises in process reengineering is the main direction of efforts to enhance the production competitiveness of enterprises. For the entire process of production and machines, if the machine and product status can be monitored in real time during the production process, and early warnings can be issued before product defects occur, it will be able to effectively reduce the cost waste in the work process, and adjust the processes and machines where problems occur, effectively improve the production process, and achieve higher results in the management of the finished products and their quality.

[0004] In view of this, the existing quality inspection method is difficult to ensure real-time quality monitoring during the production process, and it is difficult to analyze the causes of quality defects. In view of this, the inventor of the present invention has thought about and designed a quality cause analysis system to improve the deficiencies of the existing technology and further promote its implementation and utilization in the industry. Summary of the invention

[0005] In view of the above problems of the prior art, the purpose of the present invention is to provide a quality cause analysis system to solve the problem that it is difficult to find product specification defects in real time during the existing quality inspection, and it is difficult to analyze the causes of the defects and make adjustments in real time.

[0006] According to one purpose of the present invention, a quality cause analysis system is proposed, which includes an input device, a storage device, a processing device and an output device. The input device is connected to a production machine and a measuring machine, and receives the machine status data of the production machine and the product measurement data of the measuring machine. The storage device is connected to the input device, and the storage device stores the machine status data, the product measurement data and a plurality of algorithms. The processing device is connected to the storage device, and the processing device executes a plurality of control instructions to access the storage device to establish a cause analysis model, and the cause analysis model includes a quality prediction module, a network interpretation module and an optimal machine prediction module. The quality prediction module inputs the machine status data and the product measurement data into a real-time quality prediction network, and predicts whether the product measurement data meets the quality inspection regulations through the machine status data. The network interpretation module performs approximate calculations of the real-time quality prediction network through an explainable artificial intelligence algorithm (Explainable AI), and measures the contribution degree value (Shapley value) of a plurality of state variables in the machine status data to the product measurement data. The optimal machine prediction module calculates the optimal machine evaluation value (Capability of accuracy value, CK value) of the machine state data corresponding to the product measurement data, and selects a plurality of machine state data whose optimal machine evaluation value is closest to the specification center value as the optimal machine data. The output device is connected to the processing device and the storage device, and outputs the contribution degree values ​​of the plurality of state variables and the optimal machine data to analyze the quality causes of the production products.

[0007] Optionally, the production machine may include a punching machine, and the measuring machine may include an instantaneous tester.

[0008] Optionally, the real-time status data of the machine may include a data set number, a machine rotation speed, a machine status, and a number of punches per second.

[0009] Optionally, the product specification measurement data may include a data set number, an inspection time, and a measurement specification measurement value.

[0010] Optionally, the optimal machine evaluation value may include a maximum specification evaluation value and a minimum specification evaluation value of the predicted product specifications. If the maximum specification evaluation value is greater than the upper tolerance limit, the predicted product specifications are adjusted based on the maximum specification evaluation value; if the minimum specification evaluation value is less than the lower tolerance limit, the predicted product specifications are adjusted based on the minimum specification evaluation value.

[0011] Optionally, the cause analysis model may further include a recommended machine data module, which performs weighted calculation on the machine data of a preset number of periods among the optimal machine data by using the contribution value, and then calculates the correlation coefficient with the actual machine data, and selects the data with the correlation coefficient closest to 1 as the recommended machine data.

[0012] Optionally, the weighted calculation may be performed by adding the machine data to the current contribution value, where the current contribution value is the sum of the absolute values ​​of the maximum specification evaluation value and the minimum specification evaluation value.

[0013] Optionally, the cause analysis model may further include a Taguchi method analysis module to calculate system parameters in the real-time quality prediction network and obtain the best parameter combination through system parameter optimization.

[0014] Optionally, the system parameters may include a signal to noise ratio (SN).

[0015] Optionally, the signal-to-noise ratio can be calculated by multiplying the machine data of each period by the contribution value of each period and summing them up. The results are divided into two groups: within the standard deviation or outside the standard deviation. The specification values ​​of the parameter combinations are arranged in an orthogonal table, and the signal-to-noise ratio is calculated using the Nominal the best (NTB) method.

[0016] As described above, the quality cause analysis system according to the present invention may have one or more of the following advantages:

[0017] (1) This quality cause analysis system can predict product specifications based on the real-time status data of production machines by establishing a quality prediction model. During the production process, the product production status is monitored in real time. When the product specifications exceed the set rules, a real-time warning is issued to adjust the production machine to avoid causing the entire batch of defective products to be scrapped and increase production costs.

[0018] (2) This quality cause analysis system can determine the degree to which various parameters in the machine status affect quality by combining explainable artificial intelligence calculations, and obtain the most suitable machine data by analyzing the contribution of various parameters. Further obtaining the most suitable machine data allows operators to improve quality control effects by adjusting or correcting machine parameters.

[0019] (3) This quality cause analysis system can obtain optimized parameter combinations through the Taguchi method, and optimize the machine process through the data from these combinations, thereby improving overall production efficiency, reducing the time and cost required for process optimization, and improving the company's R&D and production competitiveness. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] In order to make the technical features, contents and advantages of the present invention and the effects that can be achieved more obvious, the present invention is described in detail in the form of embodiments in conjunction with the accompanying drawings as follows:

[0021] Figure 1 FIG. 4 is a block diagram of a quality cause analysis system according to an embodiment of the present invention.

[0022] Figure 2FIG. 4 is a schematic diagram of machine status data and product measurement data according to an embodiment of the present invention.

[0023] Figure 3 4 is a system architecture diagram of a quality prediction model according to an embodiment of the present invention.

[0024] Figure 4 Schematic diagram of a network interpretation module according to an embodiment of the present invention.

[0025] Figure 5 FIG. 4 is an operation flow chart of the machine data recommendation module according to an embodiment of the present invention.

[0026] Figure 6 The following is an operation flow chart of the Taguchi method analysis module according to an embodiment of the present invention.

[0027] Description of Reference Numerals

[0028] 10: Real-time quality prediction system 11: Input device

[0029] 12: Storage device 13: Processing device

[0030] 14: Output device 20,31: Input data set

[0031] 20A: Subset 21,25: Dataset number

[0032] 22: Machine speed 23: Machine status

[0033] 24: Punching times per second 26: Inspection time

[0034] 27: Measurement specification measurement value 32: Upper and lower bounds rule encoder

[0035] 33: Good product consistency rule encoder 34: Data encoder

[0036] 35: Converter Model 36: Output Data

[0037] 91: Production machine 91A: Stamping machine

[0038] 92: Measuring machine 92A: Instantaneous tester

[0039] 121: Machine status data 122: Product measurement data

[0040] 123: Algorithm 131: Cause Analysis Model

[0041] 132: Quality prediction module 133: Network interpretation module

[0042] 134: Optimal Machine Prediction Module S11~S14, S21~S24: Steps DETAILED DESCRIPTION

[0043] In order to understand the technical features, contents and advantages of the present invention and the effects that can be achieved, the present invention is hereby described in detail in the form of embodiments in conjunction with the accompanying drawings. The drawings used therein are only for illustration and auxiliary description purposes, and may not be the actual proportions and precise configurations after the implementation of the present invention. Therefore, the proportions and configuration relationships of the attached drawings should not be interpreted or limited to the scope of rights of the present invention in actual implementation. This should be stated in advance.

[0044] See also Figure 1 , which is a block diagram of a quality cause analysis system according to an embodiment of the present invention. As shown in the figure, the quality cause analysis system 10 includes an input device 11, a storage device 12, a processing device 13, and an output device 14. The input device 11 is connected to a production machine 91 and a measuring machine 92. The machine status data 121 of the production machine 91 and the product measurement data 122 of the measuring machine 92 are transmitted to the storage device 12 through the input device 11. The input device 11 can be a data collector, which is arranged on the production machine 91 and the measuring machine 92 to monitor the production status of the production machine 91 and the detection status of the measuring machine 92. In other embodiments, the input device 11 can be a data transmission interface, which is connected to the production machine 91 and the measuring machine 92. The detection data of the machine is transmitted to the storage device 12 through the data transmission interface.

[0045] The storage device 12 is connected to the input device 11, and stores machine status data 121, product measurement data 122, and a plurality of algorithms 123. The storage device 12 can be a storage medium such as a read-only memory, flash memory, disk, or cloud database of a computer device, or can be a database of a server or cloud device. When the machine status data 121 and product measurement data 122 from the input device 11 are received, they can be stored in the storage device 12 for subsequent analysis, and various analysis algorithm programs are also stored in the storage device 12. The processing device 13 is connected to the storage device 12, and the processing device 13 includes a central processing unit, a microprocessor, an image calculation processor, etc. in a computer or server device. The processing device 13 can execute control instructions to access the data in the storage device 12, and establish a cause analysis model 131 by executing calculation and analysis programs.

[0046] The cause analysis model 131 includes a quality prediction module 132, a network interpretation module 133 and an optimal machine prediction module 134. The quality prediction module 132 inputs the machine state data 121 and the product measurement data 122 into the real-time quality prediction network, and predicts whether the product measurement data 122 meets the quality inspection regulations through the machine state data 121. The network interpretation module 133 performs approximate calculations of the real-time quality prediction network through an interpretable artificial intelligence algorithm, and measures the contribution values ​​of the plurality of state variables in the machine state data 121 to the product measurement data 122. The optimal machine prediction module 134 calculates the optimal machine evaluation value of the machine state data corresponding to the product measurement data, and selects the plurality of machine state data whose optimal machine evaluation value is closest to the specification center value as the optimal machine data. The output device 14 is connected to the processing device 13 and the storage device 12, and outputs the contribution values ​​of the plurality of state variables and the optimal machine data to analyze the quality causes of the production products.

[0047] See also Figure 2 , which is a schematic diagram of the machine status data and product measurement data of an embodiment of the present invention. As shown in the above embodiment, the machine status data 121 and product measurement data 122 of the production machine 91 and the measuring machine 92 respectively represent the production status of the production machine 91 and the relevant specifications of the finished product detected by the measuring machine 92. The existing quality management mainly detects the specifications of the finished product by machine or manual after the machine completes the whole batch of products. This method is not only inefficient, but also causes a large amount of cost loss when the defective rate is high. If the product quality can be monitored in real time during the production process, it can be predicted and the production line personnel can be reminded to make corrections before the defective products occur, which can not only reduce the waste of scrapped products, but also effectively improve the production quality and yield. As shown in the figure, the production machine 91 takes the stamping machine 91A as an example, and the measuring machine 92 uses the instantaneous tester 92A to replace the manual inspection method. By real-time monitoring of the data of the stamping machine 91A and the instantaneous tester 92A, real-time quality prediction is performed to achieve the effect of real-time monitoring and management.

[0048] In the present embodiment, the machine status data 121 of the punching machine 91A includes a data set number 21, a machine speed 22, a machine status 23, and a punch count 24 per second. The product measurement data 122 of the instantaneous tester 92A includes a data set number 25, an inspection time 26, and a measurement specification measurement value 27. The data set number 21 of the machine status data 121 includes the punching time (inspection work order establishment time), the machine number, and the work order number of the punching machine 91A, and the corresponding data set number 25 of the product measurement data 122 includes the inspection time (inspection work order establishment time), the machine number, and the work order number of the instantaneous tester 92A. The machine speed 22 refers to the operating speed of the machine motor, and the machine status 23 includes the state types such as shutdown, standby, normal operation, and manual adjustment. The punch count 24 per second is the number of punches per second of the machine. The measurement specification measurement value 27 is the maximum and minimum values ​​of each specification measurement in the sampling result when the product is sampled.

[0049] By pairing the data with the same data set number (21 and 25), the input data set 20 of the machine status data 121 and the product measurement data 122 is formed, which is then used as the training data of the network model. In this embodiment, among the machine real-time status data 121 and the product specification measurement data 122 with the same data set number (21 and 25), the preset time difference between the detection time of the instantaneous tester 92A and the production time of the stamping machine 91A can be set, and the machine real-time status data 121 with the smallest difference from the preset time difference is selected to pair to form the input data set 20. For example, if the stamping production and inspection time differ by about one hour, the preset time difference can be set to one hour, and the machine real-time status data 121 and the product specification measurement data 122 one hour before the detection time point are selected to pair to form the input data set 20.

[0050] In another embodiment, bootstrap aggregating can be further used to extract samples from the training data with replacement, that is, in addition to the paired time point data, multiple periods of machine real-time status data 121 are continuously extracted from the time point with the smallest difference from the preset time difference to generate a subset 20A of the input data set 20 as model training data. These input data sets 20 can be divided into training data and test data according to a predetermined ratio, for example, divided in a ratio of 7:3. During training, the training data can be divided again to evaluate the performance of the training in a cross-validation manner.

[0051] See also Figure 3, which is a system architecture diagram of the quality prediction model of an embodiment of the present invention. As shown in the figure, after obtaining the input data set 31, the quality prediction module 132 establishes a hybrid model architecture through a plurality of algorithms 123. The hybrid model architecture includes a rule encoder, and the rule encoder includes an upper and lower bound rule encoder 32, a good product consistency rule encoder 33 and a data encoder 34. The three encoders are combined to extract the feature values ​​of the hidden layer in the neural network. The data encoder 34 refers to the predicted product specifications, and its objective function is the difference between the predicted product rules and the actual measured specifications. This difference is represented by the residual e3 of the loss function. In this embodiment, the loss function of the data encoder 34 can be represented by the mean square value of the difference between the predicted value and the actual value, as shown in equation (1).

[0052]

[0053] Among them, the actual value of y is is the predicted value.

[0054] In this embodiment, in addition to the residual between the actual value and the predicted value, the logic control rules of the upper and lower bounds rule (minmaxrule) and the good consistency rule (outbound rule) are added to improve the accuracy of the overall prediction. In the upper and lower bounds rule encoder 32, it is determined whether the maximum value of each predicted specification is greater than the minimum value of the predicted specification. If not (the maximum value of the predicted specification is less than the minimum value of the predicted specification), the upper and lower bounds rule encoder 32 generates a first loss value e1 through the upper and lower bounds rule loss function. The upper and lower bounds rule loss function of the upper and lower bounds rule encoder 32 can be expressed by equation (2).

[0055]

[0056] Where i is the predicted minimum value, j is the predicted maximum value, and n is the number of predicted specifications. Divide the predicted minimum value by the maximum value and discard it unconditionally. If the predicted minimum value is greater than the maximum value, return 1. Finally, add the results of all specifications to become the first loss value e1.

[0057] In the good product consistency rule encoder 33, it is determined whether the actual value and the predicted value of each specification are within the upper and lower limit standards. If not (the actual value of the specification meets the standard, but the predicted value of the specification does not meet the standard; the actual value of the specification does not meet the standard, but the predicted value of the specification meets the standard), a second loss value e2 is generated through the good product consistency rule loss function. The good product consistency rule loss function of the good product consistency rule encoder 33 can be expressed by equation (3).

[0058]

[0059] Among them, n is the number of predicted specifications, std is the range of the upper and lower limits of the specification, first divide the true value and the predicted value by the boundary of the specification, and unconditionally discard, so that the value greater than the specification range is converted to 1, and the value less than the specification range is converted to 0. Finally, the two values ​​representing the converted true value and predicted value are subtracted and the absolute value is taken. If the two values ​​are the same, the second loss value e2 is 0, otherwise, if the two values ​​are different, the second loss value e2 is 1.

[0060] Finally, the three loss functions are combined to adjust the weights during model training. The first loss value may include the first weight, and the second loss value may include the second weight. The residual e between the predicted value and the true value is calculated and expressed as Equation (4).

[0061] L Total =α*L minmax rule +β*L outbound rule +(1-α-β)*L Task (α∈[0,1],β∈[0,1]∧(α+β)≤1.0) (4)

[0062] α is the first weight, β is the second weight, the first weight may range from 0 to 0.4, and the second weight may range from 0 to 0.6. In other embodiments, the first weight may be 0.1, and the second weight may be 0.1.

[0063] The above data is input into the transformer model 35, which includes an encoder and a decoder. After the data is positionally encoded, the correlation between words is calculated through self-attention, that is, each tag in the context generates an embedding, which contains information about the tag itself and information about other related tags weighted by the attention weight. Multi-head attention calculates multiple attention weights between all tags, and after skipping sublayers (Residual) and normalization, enters the feedforward neural network (Feed Forward) and uses ReLU as the activation function for learning. After entering the decoder, the masked multi-head attention (Masked Multi-head Attention) is used. Since the decoder predicts in sequence, it is not necessary to calculate the correlation between the word and the subsequent word (future data) during prediction, so the information after the position is covered by a mask. Multi-head attention calculates the correlation between the output and the input, and then enters the feedforward neural network (Feed Forward) and uses ReLU as the activation function for learning. Jumping and normalization calculations may be included between these operations, and the final output data 36 is finally output. The maximum and minimum predicted values ​​output by the model are judged according to the size specifications. If the maximum predicted value or the minimum predicted value of one size code exceeds the upper and lower limits of the size specifications, it is considered a defective product and marked as 0; on the contrary, if all meet the specifications, it is a good product and marked as 1. The true value is also judged in the same way to see whether it meets the specifications and marked as a basis for accuracy calculation. In this embodiment, the self-attention converter model is used as the real-time quality prediction model of the quality prediction module 132, but the present disclosure is not limited to this. In other embodiments, other deep learning network architectures can also be selected as the real-time quality prediction model of the quality prediction module 132, such as a bidirectional long short-term memory network model.

[0064] See also Figure 4 , which is a schematic diagram of the network interpretation module of an embodiment of the present invention. After the input data set is calculated through the real-time quality prediction model and the data is output, the network interpretation module 133 can use interpretable artificial intelligence technology to perform approximate calculations of the back propagation of the quality prediction model, and then measure the contribution of each input variable to the prediction result. In this embodiment, the SHAP (Shapley Additive exPlanations) method is used to estimate the contribution value (Shapley value). Consider a feature input vector x, defined as x=[x 1 ,x 2 ,...,x D], D is the feature dimension. The contribution value of a specific feature x can be calculated by observing the change in the predicted value output by the neural network when the feature exists and does not exist. d In order to avoid retraining the neural network for every combination of existing and non-existing features, the non-existing features are replaced with approximate expected values, which are 1 if they are the values ​​of the original samples and 0 if they are replaced with randomly sampled sample values. The model is calculated as shown in Equation (5).

[0065]

[0066] Among them, φ 0 is the mean of the predicted values, φ d is the contribution value, x d ∈{0,1},x (i) For a certain sample.

[0067] By providing a small amount of training data as a reference value for sample distribution, the contribution of each input in the neural network model is quickly approximated, and its absolute value (regardless of the positive and negative contribution) is taken, and then reshaped in units of individual variables, so as to understand the contribution of variables in the machine status data 121, such as speed (machine speed), frequency (number of punches per second) and machine status, to the prediction of product measurement data in the overall input data. Figure 4 As shown in the figure, among the above variables, speed has the highest contribution to the prediction of the entire model, while frequency and state have significantly lower contributions. If the contribution of different periods is considered, the data of the first and last periods have the highest contribution to the prediction.

[0068] After obtaining the contribution degree of the variables in the machine status data to the product measurement data, the optimal machine prediction module 134 calculates the optimal machine evaluation value (CK value) of the machine status data corresponding to the product measurement data, and selects a plurality of machine status data whose optimal machine evaluation value is closest to the specification center value as the optimal machine data. When the real-time quality prediction model predicts that defective products may be produced in the future, the on-site operator can adjust the machine accordingly. The calculation method of the optimal machine evaluation value (CK value) is shown in equation (6).

[0069] CK=[detail-(U+L) / 2] / [(UL) / 2] (6)

[0070] Among them, detail is the predicted maximum and minimum values ​​of the specification, U is the upper limit of the tolerance specification, and L is the lower limit of the tolerance specification. Take the M sets of data with the most suitable machine evaluation value closest to 0 as the benchmark, as the combination of the best machine data (the lower the most suitable machine evaluation value, the closer it is to the center value of the specification), which means that these M sets of machine data are in a process state that can stably produce good products.

[0071] The optimal machine evaluation value may include the maximum specification evaluation value (Max CK) and the minimum specification evaluation value (Min CK) of the predicted product specifications. If the maximum specification evaluation value is greater than the upper tolerance limit, the predicted product specifications are adjusted based on the maximum specification evaluation value. If the minimum specification evaluation value is less than the lower tolerance limit, the predicted product specifications are adjusted based on the minimum specification evaluation value so that the prediction result can be closer to the specification center.

[0072] See also Figure 5 , which is an operation flow chart of the suggested machine data module of an embodiment of the present invention. The cause analysis model 131 may further include a suggested machine data module, as shown in the figure, which includes the following steps (S11-S14):

[0073] Step S11: Obtain the optimal machine data. Calculate the optimal machine evaluation value for the predicted maximum and minimum values ​​of N specifications respectively, and take the absolute value. For the specifications predicted to be abnormal, add and average the maximum and minimum specification evaluation values, and take the M records closest to 0 as the benchmark.

[0074] Step S12: The machine data of the preset period in the optimal machine data is weighted by the contribution value. The machine data of the previous P periods of the current machine data (actual value) and the M data (recommended value) are obtained, and weighted by the contribution value. The current contribution value is calculated by adding the absolute value of the maximum specification evaluation value and the minimum specification evaluation value.

[0075] Step S13: Calculate the correlation coefficient with the actual machine data. Calculate the correlation coefficient for each of the actual value and the machine data of the first P periods of the recommended value.

[0076] Step S14: Select the data with the correlation coefficient closest to 1 as the recommended machine data. The data with the correlation coefficient closest to 1 represents the closest sequence between the two, and the last S periods of this data are used as the most suitable machine data for prediction. The number or period of N, M, P, S can be adjusted appropriately according to the type of product or machine.

[0077] See also Figure 6 , which is an operation flow chart of the Taguchi method analysis module of an embodiment of the present invention. The cause analysis model 131 may further include a Taguchi method analysis module, calculate the system parameters in the real-time quality prediction network, and obtain the best parameter combination through system parameter optimization. As shown in the figure, it includes the following steps (S21-S24):

[0078] Step S21: Divide the parameters of the process key factors into speed, frequency, and machine status. The process key factors mainly consider the machine status data, including speed (machine rotation speed), frequency (number of punches per second), and machine status.

[0079] Step S22: Multiply the parameter of each period by the contribution value and add them up, and classify the calculated results into two groups: within one standard deviation or outside one standard deviation. Taking the orthogonal table in Table 2 below as an example, the parameter combinations in the orthogonal table are divided into within or outside X standard deviations, represented by 1 and 2.

[0080] Table 2

[0081]

[0082] Step S23: Arrange the combinations according to the orthogonal table and find the results (y values) calculated by these combinations. The y value is the specification value predicted by the parameter combination. is the average of the two y values ​​(y1, y2).

[0083] Step S24: The result is put into the formula of the eye-catching characteristic in the Taguchi method to calculate the signal-to-noise ratio, thereby finding the optimal parameters. The formula of the eye-catching characteristic is expressed as equations (7) and (8).

[0084]

[0085] SN=10*log 10 (y 2 / s 2 ) (8)

[0086] The Taguchi method is a method of optimizing system parameters through experiments. Different combinations of parameter levels will result in different specifications for the final product. Through parameter design, a set of optimal parameter level combinations can be found to minimize various variations, making the product least sensitive to the source of variation. The key factors that determine the process can be obtained by using the Wangmu formula in the Taguchi method. When yield problems occur in the future, this key factor can be dealt with first.

[0087] The above description is for illustrative purposes only and is not intended to be limiting. Any equivalent modifications or changes made thereto without departing from the spirit and scope of the present invention shall be included in the scope of the appended claims.

Claims

1. A quality cause analysis system, comprising: An input device connected to the production machine and the measuring machine to receive machine status data of the production machine and product measurement data of the measuring machine; A storage device connected to the input device, wherein the storage device stores the machine status data, the product measurement data and a plurality of algorithms; A processing device is connected to the storage device, and the processing device executes a plurality of control instructions to access the storage device to establish a cause analysis model, wherein the cause analysis model includes the following modules: A quality prediction module, which inputs the machine status data and the product measurement data into a real-time quality prediction network, and predicts whether the product measurement data meets the quality inspection regulations based on the machine status data; A network interpretation module, which performs approximate calculations of the real-time quality prediction network through an interpretable artificial intelligence algorithm to measure the contribution of multiple state variables in the machine state data to the product measurement data; as well as An optimal machine prediction module calculates the optimal machine evaluation value of the machine status data corresponding to the product measurement data, and selects a plurality of machine status data whose optimal machine evaluation value is closest to the specification center value as the optimal machine data; as well as The output device is connected to the processing device and the storage device, and outputs the contribution values ​​of the plurality of state variables and the optimal machine data to analyze the quality causes of the produced products. 2 . The quality cause analysis system as claimed in claim 1 , wherein the production machine comprises a stamping machine, and the measuring machine comprises an instantaneous measuring instrument. 3 . The quality cause analysis system as claimed in claim 2 , wherein the real-time status data of the machine comprises a data set number, a machine speed, a machine status and a number of punches per second. 4 . The quality cause analysis system of claim 2 , wherein the product specification measurement data comprises a data set number, an inspection time, and a measurement specification measurement value.

5. The quality cause analysis system as claimed in claim 1, wherein the optimal machine evaluation value comprises a maximum specification evaluation value and a minimum specification evaluation value of the predicted product specification, if the maximum specification evaluation value is greater than an upper tolerance limit, the predicted product specification is adjusted based on the maximum specification evaluation value, and if the minimum specification evaluation value is less than a lower tolerance limit, the predicted product specification is adjusted based on the minimum specification evaluation value.

6. The quality cause analysis system of claim 5, wherein the cause analysis model further comprises a machine data recommendation module, which performs weighted calculation on machine data of a preset number of periods among the optimal machine data by using the contribution value, and then calculates correlation coefficients with actual machine data, and selects data with a correlation coefficient closest to 1 as the recommended machine data.

7. The quality cause analysis system of claim 6, wherein weighted calculation is to add the machine data to a current contribution value, wherein the current contribution value is the absolute value of the maximum specification evaluation value and the minimum specification evaluation value.

8. The quality cause analysis system of claim 1, wherein the cause analysis model further comprises a Taguchi method analysis module to calculate system parameters in the real-time quality prediction network and obtain the best parameter combination by optimizing the system parameters.

9. The mass factor analysis system of claim 8, wherein the system parameter comprises a signal-to-noise ratio.

10. The quality cause analysis system as claimed in claim 9, wherein the signal-to-noise ratio is obtained by multiplying the machine data of each period by the contribution value of each period and summing the sum, dividing the results into two groups: within the standard deviation or outside the standard deviation, arranging the specification values ​​of the calculation parameter combination through an orthogonal table, and applying the method of visual characteristics to calculate the signal-to-noise ratio.