Crude foil engine fault early warning method and system based on deep learning, and medium

Abstract

The invention discloses a crude foil engine fault early warning method and system based on deep learning and a medium, and the method comprises the following steps: S1, building a fault early warning model of a crude foil engine through deep learning according to historical operation data; s2, periodically acquiring an operation state vector of the crude foil engine and a quality state of the produced copper foil, wherein each component of the operation state vector corresponds to one operation state index of the crude foil engine; s3, judging whether the running state vector is matched with the fault early warning model or not; s4, if the running state vectors are matched with the fault early warning model, the fault early warning model classifies the running state vectors; and S5, judging whether the quality state of the copper foil meets the standard requirement or not, if not, sending out a fault early warning prompt of a type corresponding to the operation state vector, and if yes, re-classifying the operation state vector as a normal state, and correcting the fault early warning model, so that the accuracy of a prediction result of the fault early warning model is continuously improved.

Description

Technical Field

[0001] The present invention relates to the technical field of scanning devices, and particularly relates to a method, a system and a medium for fault warning of a copper foil generator based on deep learning. Background Art

[0002] A copper foil generator is a commonly used production device for electrolytic copper foil production, which mainly includes an electrolytic cell and a cathode roller. An anode plate is arranged in the electrolytic cell and filled with electrolyte. The cathode roller is mounted on the electrolytic cell and partially contacts the electrolyte. When an electric current is applied to the anode plate and the cathode roller, a stable electric field can be generated between the anode plate and the cathode roller in the electrolyte (copper sulfate solution), so that copper ions in the copper sulfate solution are adsorbed on the cathode roller. When the copper foil generator fails, various problems will occur in the produced copper foil, such as uneven thickness, spots, oxidation marks, edge oxidation, etc. Currently, the fault judgment of the anode plate of the copper foil generator usually uses a CCD camera to inspect the appearance of the produced copper foil, and analyzes the fault type of the anode plate according to the appearance defects of the copper foil. This can only be analyzed after the copper foil generator has failed. And some phenomena on the copper foil may also be caused by the failure of the cathode roller. Therefore, this analysis method often cannot accurately give the specific location or type of the fault, and at the same time cannot reflect the working state of the copper foil generator in real time, resulting in a large number of unqualified copper foils and increasing production costs. Summary of the Invention

[0003] In order to solve the technical problems of being unable to give early warning of the faults of the copper foil generator and accurately give the fault types, the present application provides a method, a system and a medium for fault warning of a copper foil generator based on deep learning.

[0004] Among them, the method for fault warning of a copper foil generator based on deep learning includes the following steps: S1. According to historical operation data, use deep learning to establish a fault warning model of the copper foil generator; S2. Periodically obtain the operation state vector of the copper foil generator and the quality state of the produced copper foil. Each component of the operation state vector corresponds to a state index of the copper foil generator; S3. Judge whether the operation state vector matches the fault warning model; S4. If the operation state vector matches the fault warning model, the fault warning model classifies the operation state vector; S5. Judge whether the quality state of the copper foil meets the standard requirements. If not, issue a fault warning prompt of the type corresponding to the operation state vector. If so, reclassify the operation state vector as a normal state and correct the fault warning model.

[0005] Use deep learning to establish a fault warning model for the copper foil machine to predict the operating status of the copper foil machine, and at the same time check the quality status of the produced copper foil to evaluate the prediction results of the fault warning model, so as to realize the correction of the fault warning model, make the fault warning model can be continuously iterated, and improve the accuracy of the prediction results.

[0006] Specifically, the fault warning model is obtained through the following steps: S11. Obtain a plurality of historical operation data, and part of the historical operation data is fault information; S12. Create a historical operation data set according to all the historical operation data; S13. Split the historical operation data set into a training data set and a test data set; S14. Use the training data set to train the initial model; S15. Use the test data set to optimize the parameters of the initial model to obtain the fault warning model.

[0007] Specifically, the historical operation data set is obtained through the following steps: S121. Classify the historical operation data through manual expert classification and mark the sample labels; S122. Perform normalization processing on the historical operation data; S123. Construct the historical operation data set according to the sample labels.

[0008] Specifically, the historical operation data set includes a plurality of historical fault samples, each historical fault sample includes a plurality of fault characteristic values and a sample label, each fault characteristic value corresponds to a state index of the copper foil machine, and each sample label corresponds to a fault type of the copper foil machine.

[0009] Specifically, the operating state vector is obtained through the following steps: S21. Obtain a plurality of operating state indexes of the copper foil machine; S22. Divide each operating state index by its corresponding standard state value one by one, and use the obtained quotient as a component of the operating state vector.

[0010] Specifically, in step S3, the following steps are used to determine whether the operating state vector matches the fault warning model: S31. Use the fault warning model to construct a predicted fault data set, and the predicted fault data set includes a plurality of predicted fault samples; S32. Create a predicted fault vector for each of the predicted fault samples, where each component of the predicted fault vector represents the same operating status metric as the corresponding component of the operating status vector. S33. Select one of the predicted fault vectors one by one, subtract the operating status vector from the predicted fault vector to obtain a difference vector, and calculate the norm of the difference vector. S34. Compare the norm of the difference vector with a first threshold. If it is less than the first threshold, determine that the operating status vector matches the fault warning model, and mark the label corresponding to the predicted fault vector as the type of the operating status vector.

[0011] By adjusting the size of the first threshold, the difference between the operating status vector and the predicted fault vector can be changed, thereby changing the sensitivity of the fault warning, so as to achieve fault prompting before an actual fault and avoid the production of unqualified copper foil.

[0012] Specifically, the quality status of the copper foil is obtained through the following steps: S23. Periodically obtain the thickness values of multiple points on the copper foil. S24. Calculate the variance of all the thickness values. S25. Compare the variance with a second threshold. If the variance is greater than the second threshold, determine that the quality status of the copper foil does not meet the standard requirements.

[0013] The thickness values of a copper foil with qualified quality should fluctuate within a certain range. When the thickness values of the copper foil fluctuate beyond a certain range, it means that the working state of the copper foil making machine is abnormal. Therefore, fault prediction and prompting are required.

[0014] Further, before performing step S25, the following steps are first performed: S251. Periodically obtain the image information of the copper foil. S252. Use image recognition technology to analyze whether there are apparent defects on the copper foil. If there are, determine that the quality status of the copper foil does not meet the standard requirements. If not, perform step S25.

[0015] Considering that the current thickness measurement technology cannot measure the thickness of all points on the copper foil, there may be a situation of missed detection of defects. Therefore, it is necessary to use image recognition technology to re-check the quality status of the copper foil to achieve double detection.

[0016] The present application also provides a fault warning system for a copper foil machine based on deep learning, including a memory, a processor, and a deep learning-based copper foil machine fault warning program stored on the memory and executable on the processor. When the deep learning-based copper foil machine fault warning program is executed by the processor, it implements the steps of the above-mentioned deep learning-based copper foil machine fault warning method.

[0017] The present application also provides a computer-readable storage medium with a deep learning-based copper foil machine fault warning program stored thereon. When this program is executed by a processor, it implements the steps of the above-mentioned deep learning-based copper foil machine fault warning method.

[0018] Technical effects and advantages of the present invention: 1. Use deep learning to establish a fault warning model for the copper foil machine to predict the operating state of the copper foil machine, and at the same time check the quality state of the produced copper foil to evaluate the prediction results of the fault warning model, so as to correct the fault warning model, enabling the fault warning model to be continuously iterated and improving the accuracy of the prediction results; 2. By adjusting the size of the first threshold, the difference between the operating state vector and the predicted fault vector can be changed, thereby changing the sensitivity of the fault warning, so as to achieve a fault prompt before an actual fault and avoid the production of unqualified copper foil; 3. Judge the quality state of the copper foil by measuring the thickness of the copper foil and the image information of the copper foil, realizing a double detection of the quality state of the copper foil, thus avoiding misjudgment due to missed detection and improving the accuracy of the prediction results. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 It is the overall flowchart of the copper foil machine fault warning method of the present invention.

[0020] Figure 2 It is the flowchart of obtaining the fault warning model in the present invention.

[0021] Figure 3 It is the flowchart of judging whether the operating state vector matches the fault warning model in the present invention.

[0022] Figure 4 It is the flowchart of the steps for judging the quality state of the copper foil in the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0023] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0024] When the copper foil production machine produces copper foil, failures of different components will cause various defects or flaws in the produced copper foil. Merely inferring the faulty components from the defects or flaws of the copper foil often fails to obtain accurate results. Moreover, failures of different components will cause different changes in the operating state of the copper foil production machine. For example, when the anode plate is corroded, the coating peels off, or it is deformed, the current density distribution of the anode plate will be uneven, which will further lead to uneven thickness of the copper foil; when the cathode roller fails, it will cause abnormal current of the cathode roller or abnormal rotation speed of the cathode roller, etc.; when the composition of the electrolyte is abnormal, it may cause abnormal changes in the current of the electrolyte, etc. According to actual engineering experience, no failure occurs suddenly, but is caused by the continuous accumulation of some small defects or problems to a certain extent. During the accumulation process, certain state indicators of the equipment operation will change continuously. Therefore, by monitoring the operating state of the copper foil production machine and combining deep learning, it is possible to achieve fault diagnosis and early warning of the copper foil production machine and determine the faulty components.

[0025] Reference Figures 1 to 4 , an embodiment of the present application provides a fault early warning method for a copper foil production machine based on deep learning, including the following steps: S1. According to historical operation data, use deep learning to establish a fault early warning model for the copper foil production machine; S2. Periodically obtain the operation state vector of the copper foil production machine and the quality state of the produced copper foil. Each component of the operation state vector corresponds to an operation state indicator of the copper foil production machine; S3. Determine whether the operation state vector matches the fault early warning model; S4. If the operation state vector matches the fault early warning model, the fault early warning model classifies the operation state vector; S5. Determine whether the quality state of the copper foil meets the standard requirements. If it does not meet the requirements, issue a fault early warning prompt of the type corresponding to the operation state vector. If it meets the requirements, reclassify the operation state vector as the normal state and correct the fault early warning model.

[0026] Use deep learning to establish a fault early warning model for the copper foil production machine to predict the operating state of the copper foil production machine, and at the same time check the quality state of the produced copper foil to evaluate the prediction results of the fault early warning model, so as to correct the fault early warning model, enable the fault early warning model to be continuously iterated, and improve the accuracy of the prediction results.

[0027] Specifically, reference Figure 2 , in step S1, the fault early warning model is obtained through the following steps: S11. Obtain a plurality of historical operation data, and some of the historical operation data are fault data; S12. Create a historical operation dataset based on all historical operation data; S13. Split the historical operation dataset into a training dataset and a test dataset; S14. Use the training dataset to train the initial model; S15. Use the test dataset to optimize the parameters of the initial model to obtain a fault warning model.

[0028] The splitting of the historical operation dataset can be done by the hold-out method into two groups, one is the training dataset and the other is the test dataset. It can also be done by the K-fold cross-validation method, where the historical operation dataset is split into multiple sub-training sets. Each time, one of the sub-training sets is used as the test dataset to correct the fault warning model, and the other sub-training sets are used as the training dataset to train the fault warning model, thus achieving cross-validation of the output results of the fault warning model to improve the accuracy of the fault warning model.

[0029] In step S12, the historical operation dataset can be created manually or by automatic feature extraction and automatic classification using deep learning algorithms. When there is a large amount of historical fault data, manual experts pre-classify some of the data, and then the deep learning algorithm is used for automatic classification. The specific steps are as follows: S121. Classify the historical operation data through manual expert classification and mark the sample labels; S122. Perform normalization processing on the historical operation data; S123. Construct a historical operation dataset according to the sample labels.

[0030] The historical operation dataset includes multiple historical fault samples. Each historical fault sample includes multiple fault feature values and a sample label. Each fault feature value corresponds to a state index of the copper foil generator respectively, and each sample label corresponds to a fault type of the copper foil generator respectively.

[0031] Of course, the historical operation dataset can also include normal operation state samples under the normal operation state of the copper foil generator. The normal operation state samples also participate in the training of the fault warning model, which can effectively improve the accuracy of the output results of the fault warning model.

[0032] The following table shows part of the content of the historical operation dataset, which is used to illustrate the composition of the historical operation dataset, but not a limitation on the historical operation dataset. Key Fl_Sol Te_Sol Co_Sol Hi_Sol I_Pos Sp_Neg Class Label 1 0.954 0.990 1.112 1.010 1.752 1.327 Pos_Fault 2 0.955 1.011 1.091 0.995 1.332 1.891 Neg_Fault 3 1.009 1.156 1.727 0.998 1.791 1.824 Sol_Abnor 4 0.789 1.329 1.023 0.769 1.515 1.623 Sol_Leak 5 0.979 1.015 1.101 0.967 1.023 0.998 All_Norm

[0033] In the above table, Key represents the primary key of each historical fault sample. Fl_Sol, Te_Sol, Co_Sol, Hi_Sol, I_Pos, and Sp_Neg respectively correspond to an operating status indicator of the raw copper foil machine, which are the flow rate of the liquid supply circuit, the temperature of the electrolyte, the composition of the electrolyte, the liquid level height, the current of the anode plate, and the rotational speed of the cathode roller. The values in the columns corresponding to the respective operating status indicators are the weighted values obtained by calculating the historical operating data.

[0034] Class Label represents the sample label. Different sample labels correspond to different fault types or the operating status of the raw copper foil machine. Pos_Fault represents anode plate fault, Neg_Fault represents cathode roller fault, Sol_Abnor represents abnormal electrolyte composition, Sol_Leak represents electrolyte leakage, and All_Norm represents normal operating status.

[0035] The historical operating data set can also include more fault characteristic values to record more operating status data of the raw copper foil machine.

[0036] Generally, in order to reduce each operating status indicator, normalization processing is required when constructing the operating status vector to reduce the impact of different dimensions of the actual values on model training. In the first embodiment of the present application, the operating status vector is normalized through the following steps: S21. Obtain the operating status indicators of multiple raw copper foil machines; S22. Divide each operating status indicator by its corresponding standard status value one by one, and use the obtained quotient as the component of the operating status vector.

[0037] The standard status value means that when the raw copper foil machine is operating normally, each operating status indicator should fluctuate around a fixed value, and this fixed value can be used as the standard value of this operating status indicator.

[0038] Such a normalization processing method has less calculation amount, and the obtained components can intuitively determine the deviation degree between the actual operating status indicator and the standard value, which is convenient for subsequent calculations.

[0039] Reference Figure 3 , in step S3, it is judged whether the operating status vector matches the fault warning model through the following steps: S31. Use the fault warning model to construct a predicted fault data set, and the predicted fault data set includes multiple predicted fault samples; S32. Create a predicted fault vector for each predicted fault sample. Each component of the predicted fault vector represents the same operating status indicator as the component of the corresponding operating status vector; S33. Select a predicted fault vector one by one, subtract the operating state vector from the predicted fault vector to obtain a difference vector, and calculate the norm of the difference vector. S34. Compare the norm of the difference vector with the first threshold. If it is less than the first threshold, it is determined that the operating state vector matches the fault warning model, and the label corresponding to the predicted fault vector is marked as the type of the operating state vector.

[0040] The structure form of the predicted fault data set generated by the fault warning model is the same as the above table. The difference is that each fault type contains a predicted fault sample, and according to the fault characteristic values recorded in each predicted fault sample, the corresponding predicted fault vector is formed. Taking the above table as an example, four predicted fault vectors are created for the four fault labels: Pos_Fault = [0.954, 0.990, 1.112, 1.010, 1.752, 1.327]; Neg_Fault = [0.955, 1.011, 1.091, 0.995, 1.332, 1.891]; Sol_Abnor = [1.009, 1.156, 1.727, 0.995, 1.791, 1.824]; Sol_Leak = [0.789, 1.329, 1.023, 0.769, 1.515, 1.623]; When the fault warning model creates a predicted fault vector, the dimension of the predicted fault vector should be consistent with the operating state vector, and the meaning represented by each component should be the same, so that the two can be compared. To judge whether two vectors are the same, it is mainly through judging the included angle and the norm of the two vectors. Therefore, the difference between the operating state vector and the predicted fault vector can be used to judge the approximation degree of the two vectors.

[0041] By modifying the size of the first threshold, the sensitivity of the fault warning can be changed. The larger the first threshold, it means that the difference between the operating state vector and the predicted fault vector is greater when a fault prompt is made, which means that the fault warning is more sensitive. When the first threshold is smaller, the difference between the operating state vector and the predicted fault vector is smaller when a fault prompt is made. Of course, the smaller the first threshold means that the fault warning model is more likely to misjudge a fault and wrongly issue a fault warning prompt. Therefore, the setting of the first threshold should be adjusted according to the actual operating state of the copper foil generator.

[0042] The measurement of the quality status of copper foil is currently mostly achieved by combining a CCD camera with image processing technology. However, this can only detect defects visible to the naked eye, such as wrinkles, pinholes, spots, etc. At this time, the quality of the copper foil no longer meets the standard requirements. Therefore, it is necessary to adopt a more accurate measurement method to detect problems before serious defects occur in the copper foil. The laser dot matrix technology can achieve this well. The laser dot matrix technology measures the thickness of multiple points on the copper foil quickly and with high precision through high-frequency lasers. When the raw foil machine is in normal operation, the thickness of the produced copper foil fluctuates around a certain fixed value, and its fluctuation range is controlled within a reasonable interval. When the operating state of the raw foil machine is abnormal, the fluctuation range of the copper foil thickness will abnormally expand. Therefore, referring to Figure 4 , the quality status of the copper foil is obtained through the following steps: S23. Periodically obtain the thickness values of multiple points on the copper foil; S24. Calculate the variance of all thickness values; S25. Compare the variance with the second threshold. If the variance is greater than the second threshold, it is determined that the quality status of the copper foil does not meet the standard requirements.

[0043] The judgment on whether the quality status of the copper foil meets the standard requirements can also change the accuracy of the result by changing the judgment criteria, so as to achieve early warning before a failure. For example, in the above steps, by changing the size of the second threshold, the strictness of the control of the quality status of the copper foil can be changed. If the second threshold is reduced, it means that the fluctuation range of the copper foil thickness is narrowed, the sensitivity of the failure early warning is improved, and the failure early warning model issues a failure early warning prompt earlier. Of course, this also means that the failure early warning model is more likely to give false alarms. Therefore, the value of the second threshold should be adjusted according to the actual operating state of the raw foil machine.

[0044] Of course, when determining the quality of the copper foil by the copper foil thickness, the maximum thickness value or the minimum thickness value can also be used for determination, but this method has a relatively high false judgment rate for failures.

[0045] Considering that when the laser dot matrix measures the copper foil thickness, it cannot fully cover the entire copper foil and there is a possibility of missing defects, an image recognition method can be adopted to further ensure the accuracy of the detection result.

[0046] Specifically, referring to Figure 4 , before executing step S25, the following steps are first executed: S251. Periodically obtain the image information of the copper foil; S252. Use image recognition technology to analyze whether there are apparent defects on the copper foil. If there are, it is determined that the quality status of the copper foil does not meet the standard requirements. If not, step S25 is executed.

[0047] The specific technical details of implementing image recognition technology are currently relatively mature and have been applied in some projects, so they will not be elaborated here.

[0048] This application also provides a fault warning system for a copper foil machine based on deep learning, including a memory, a processor, and a copper foil machine fault warning program based on deep learning stored on the memory and executable on the processor. When the copper foil machine fault warning program based on deep learning is executed by the processor, the steps of the above-mentioned copper foil machine fault warning method based on deep learning are implemented.

[0049] This application also provides a computer-readable storage medium with a copper foil machine fault warning program based on deep learning stored thereon. When this program is executed by the processor, the steps of the above-mentioned copper foil machine fault warning method based on deep learning are implemented.

[0050] The specific implementation manners of the above system and the readable storage medium are well-known technologies to those skilled in the art, so they will not be elaborated here.

[0051] Finally: The above are only the preferred embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. A foil machine fault warning method based on deep learning, characterized in that: The following steps are involved: S1. Based on historical operation data, a fault warning model of the foil machine is established using deep learning; S2, periodically obtaining the operation state vector of the foil machine and the quality state of the produced copper foil, wherein each component of the operation state vector corresponds to an operation state indicator of the foil machine; S3, judging whether the operation state vector matches the fault warning model; S4. If the operation state vector matches the fault warning model, the fault warning model classifies the operation state vector; S5. Determine whether the quality status of the copper foil meets the standard requirements. If not, issue a fault warning prompt of the type corresponding to the operating state vector. If yes, reclassify the operating state vector as a normal state and modify the fault warning model.

2. The method for early warning of foil production machine failure based on deep learning according to claim 1 is characterized in that: The fault warning model is obtained by the following steps: S11, obtaining a plurality of historical operation data, wherein some of the historical operation data are fault data; S12, creating a historical operation data set according to all the historical operation data; S13, splitting the historical operation data set into a training data set and a test data set; S14, training the initial model using the training data set; S15. Optimize the parameters of the initial model using the test data set to obtain the fault warning model.

3. The method for early warning of foil production machine failure based on deep learning according to claim 2 is characterized in that: The historical operation data set is obtained by the following steps: S121, classifying the historical operation data through manual expert classification and marking sample labels; S122, normalizing the historical operation data; S123. Construct the historical operation data set according to the sample labels.

4. The method for early warning of foil production machine failure based on deep learning according to claim 3 is characterized in that: The historical operation data set includes multiple historical fault samples, each of which includes multiple fault feature values ​​and a sample label, each of which corresponds to a state indicator of the foil production machine, and each of which corresponds to a fault type of the foil production machine.

5. The method for early warning of foil production machine failure based on deep learning according to claim 1, characterized in that: The running state vector is obtained by the following steps: S21, obtaining a plurality of operating status indicators of the foil production machine; S22. Divide each of the operating status indicators by its corresponding standard status value one by one, and use the obtained quotients as components of the operating status vector.

6. The method for early warning of foil production machine failure based on deep learning according to claim 5 is characterized in that: In step S3, it is determined whether the operating state vector matches the fault warning model by the following steps: S31, constructing a predicted fault data set using the fault warning model, wherein the predicted fault data set includes a plurality of predicted fault samples; S32, creating a predicted fault vector for each predicted fault sample, wherein each component of the predicted fault vector and the corresponding component of the running state vector represent the same running state indicator; S33, selecting one of the predicted fault vectors one by one, subtracting the running state vector from the predicted fault vector to obtain a difference vector, and calculating the modulus of the difference vector; S34. Compare the modulus of the difference vector with a first threshold value. If the modulus is less than the first threshold value, determine that the operating state vector matches the fault warning model, and mark the label corresponding to the predicted fault vector as the type of the operating state vector.

7. The method for early warning of foil production machine failure based on deep learning according to claim 1, characterized in that: The quality status of the copper foil is obtained by the following steps: S23, periodically obtaining thickness values ​​of multiple points on the copper foil; S24, calculating the variance of all the thickness values; S25, comparing the variance with a second threshold value, and if the variance is greater than the second threshold value, determining that the quality state of the copper foil does not meet standard requirements.

8. The method for early warning of foil production machine failure based on deep learning according to claim 7 is characterized in that: Before executing step S25, the following steps are performed: S251, periodically acquiring image information of the copper foil; S252, using image recognition technology to analyze whether the copper foil has any surface defects, if so, determining that the quality status of the copper foil does not meet the standard requirements, if not, executing step S25.

9. A foil machine fault warning system based on deep learning, characterized in that: A foil machine fault warning system based on deep learning comprises: a memory, a processor, and a foil machine fault warning program based on deep learning stored in the memory and executable on the processor. When the foil machine fault warning program based on deep learning is executed by the processor, the steps of the foil machine fault warning method based on deep learning as described in any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium, characterized in that: A deep learning-based foil production machine fault warning program is stored on a readable storage medium, and when the program is executed by a processor, the steps of the deep learning-based foil production machine fault warning method as described in any one of claims 1 to 8 are implemented.

Images