Industrial internet of things production line data analysis method, system, device and medium

Abstract

The application discloses an industrial Internet of Things production line data analysis method, system, device and medium, relates to the technical field of the industrial Internet of Things, and the method comprises the following steps: acquiring process parameter data and equipment power consumption data corresponding to each workpiece processed by a target production line; for each workpiece, taking the workpiece as a target workpiece, performing first focused mining on target equipment power consumption data of the target workpiece based on non-target equipment power consumption data corresponding to non-target workpieces, and obtaining power consumption focusing features of the target workpiece; performing second focused mining on the power consumption focusing features based on the process parameter data, and obtaining process power consumption correlation features of the target workpiece; and analyzing corresponding workpiece quality based on the process power consumption correlation features, and obtaining workpiece quality analysis results. The application has the effect of improving the accuracy of workpiece quality analysis in a production line.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of industrial Internet of Things, and in particular to an industrial Internet of Things production line data analysis method, system, device and medium. BACKGROUND

[0002] In the field of intelligent manufacturing, Internet of Things technology has been widely applied to production lines. By deploying various sensors, real-time collection of equipment status, process parameters and energy consumption and other massive data is realized. These data contain key information reflecting the equipment operation health, process execution and final product quality. Therefore, analyzing production line data to realize real-time monitoring, prediction and traceability of product quality has become a key link to improve production efficiency and product quality.

[0003] The existing analysis method mainly has the following limitations: first, most researches focus on isolated analysis of single workpiece data, such as establishing a regression or classification model directly on the time series data of equipment power consumption or process parameters. This method ignores the internal correlation and difference between different workpiece data on the same production line, and cannot highlight the abnormal characteristics of a specific workpiece through comparison, resulting in insufficient detection sensitivity for minor defects or unstable states.

[0004] Secondly, when fusing multi-source data, the existing technology often uses simple feature splicing or shallow model for processing. However, there is a complex and nonlinear dynamic coupling relationship between process parameters and equipment power consumption. Simple fusion methods cannot deeply mine the deep semantic relationship between them, resulting in insufficient information utilization and affecting the accuracy and robustness of the quality analysis model.

[0005] Therefore, there is an urgent need for a new data analysis method to more accurately and reliably complete the intelligent analysis of workpiece quality through effective comparison mechanism and deep feature fusion technology. SUMMARY

[0006] In order to improve the accuracy of workpiece quality analysis in the production line, the present application provides an industrial Internet of Things production line data analysis method, system, device and medium.

[0007] In a first aspect, the present application provides an industrial Internet of Things production line data analysis method, which adopts the following technical solution:

[0008] The industrial Internet of Things production line data analysis method is applied to an industrial Internet of Things system, which includes a management platform, a sensor network platform and an object platform connected in sequence. The method is executed by the management platform and includes:

[0009] Obtaining process parameter data and equipment power consumption data corresponding to each workpiece processed by the target production line, wherein the process parameter data at least includes temperature time series data and pressure time series data, and the equipment power consumption data is power consumption time series data of corresponding processing equipment when the workpiece is produced;

[0010] For each workpiece, taking the workpiece as a target workpiece, based on non-target equipment power consumption data corresponding to non-target workpieces, performing first focused mining on target equipment power consumption data of the target workpiece to obtain power consumption focused features of the target workpiece;

[0011] Based on the process parameter data, performing second focused mining on the power consumption focused features to obtain process power consumption correlation features of the target workpiece;

[0012] Based on the process power consumption correlation features, analyzing corresponding workpiece quality to obtain workpiece quality analysis results.

[0013] By adopting the above technical solution, process parameter data and equipment power consumption data corresponding to each workpiece processed by the target production line are obtained, wherein the process parameter data at least includes temperature time series data and pressure time series data, and the equipment power consumption data is power consumption time series data of corresponding processing equipment when the workpiece is produced. Then, for each workpiece, taking the workpiece as a target workpiece, based on non-target equipment power consumption data corresponding to non-target workpieces, performing first focused mining on target equipment power consumption data of the target workpiece to obtain power consumption focused features of the target workpiece. Then, based on the process parameter data, performing second focused mining on the power consumption focused features to obtain process power consumption correlation features of the target workpiece. Then, based on the process power consumption correlation features, analyzing corresponding workpiece quality to obtain workpiece quality analysis results. By introducing a double focused mining mechanism, the present application effectively overcomes the defects of isolated analysis of workpiece data and insufficient multi-source data fusion in the prior art, so that the quality prediction and defect diagnosis results of the workpiece are more accurate and reliable, and strong decision support is provided for intelligent quality control and process optimization of the production line.

[0014] Optionally, the step of performing first focused mining on target equipment power consumption data of the target workpiece based on non-target equipment power consumption data corresponding to non-target workpieces to obtain power consumption focused features of the target workpiece comprises:

[0015] Performing semantic embedding on the target equipment power consumption data to obtain corresponding first power consumption embedding features, and performing semantic embedding on non-target equipment power consumption data corresponding to each non-target workpiece to obtain corresponding second power consumption embedding features;

[0016] load the first power consumption embedding feature and the second power consumption embedding feature into a first focused mining unit of a workpiece quality analysis network, wherein the first focused mining unit comprises a first power consumption focused branch and a second power consumption focused branch;

[0017] for each second power consumption embedding feature, focus mine the first power consumption embedding feature based on the second power consumption embedding feature through the first power consumption focused branch to obtain a first power consumption focused feature, and focus mine the first power consumption embedding feature based on the second power consumption embedding feature through the second power consumption focused branch to obtain a second power consumption focused feature;

[0018] splice the first power consumption focused feature, the second power consumption focused feature and the first power consumption embedding feature to obtain a power consumption focused feature of the target workpiece.

[0019] By adopting the technical scheme, in order to obtain a power consumption focused feature of a target workpiece, semantic embedding is performed on target device power consumption data to obtain a corresponding first power consumption embedding feature, and semantic embedding is respectively performed on non-target device power consumption data corresponding to each non-target workpiece to obtain a corresponding second power consumption embedding feature, then the first power consumption embedding feature and the second power consumption embedding feature are loaded into a first focused mining unit of a workpiece quality analysis network, wherein the first focused mining unit comprises a first power consumption focused branch and a second power consumption focused branch, then for each second power consumption embedding feature, the first power consumption embedding feature is focus mined based on the second power consumption embedding feature through the first power consumption focused branch to obtain a first power consumption focused feature, and the first power consumption embedding feature is focus mined based on the second power consumption embedding feature through the second power consumption focused branch to obtain a second power consumption focused feature, then the first power consumption focused feature, the second power consumption focused feature and the first power consumption embedding feature are spliced to obtain a power consumption focused feature of the target workpiece.

[0020] Optionally, the step of focus mining the first power consumption embedding feature based on the second power consumption embedding feature through the first power consumption focused branch to obtain the first power consumption focused feature comprises:

[0021] load the first power consumption embedding feature and the second power consumption embedding feature into the first power consumption focused branch, wherein the first power consumption focused branch has a first linear mapping layer, a second linear mapping layer and a third linear mapping layer;

[0022] linearly map the first power consumption embedding feature based on the first linear mapping layer to generate a query feature vector;

[0023] linearly map the second power consumption embedding feature through the second linear mapping layer to generate a key feature vector;

[0024] linearly mapping the second power consumption embedding feature based on the third linear mapping layer to generate a value feature vector;

[0025] determining a similarity matrix between the query feature vector and the key feature vector, and performing normalization processing on the similarity matrix to generate a normalized weight matrix;

[0026] performing weighted summation on the value feature vector based on the normalized weight matrix to obtain a first power consumption focusing feature.

[0027] By adopting the above technical solution, in order to obtain the first power consumption focusing feature, the first power consumption embedding feature and the second power consumption embedding feature are loaded into the first power consumption focusing branch, wherein the first power consumption focusing branch has a first linear mapping layer, a second linear mapping layer and a third linear mapping layer, then the first power consumption embedding feature is linearly mapped based on the first linear mapping layer to generate a query feature vector, then the second power consumption embedding feature is linearly mapped through the second linear mapping layer to generate a key feature vector, then the second power consumption embedding feature is linearly mapped based on the third linear mapping layer to generate a value feature vector, then the similarity matrix between the query feature vector and the key feature vector is determined, and the similarity matrix is normalized to generate a normalized weight matrix, then the value feature vector is weighted and summed based on the normalized weight matrix to obtain the first power consumption focusing feature.

[0028] Optionally, the step of focusing mining the power consumption focusing feature based on the process parameter data to obtain the process power consumption correlation feature of the target workpiece includes:

[0029] performing semantic embedding on the process parameter data to obtain corresponding process parameter embedding features;

[0030] performing semantic mining on the process parameter embedding features to obtain process parameter features corresponding to the process parameter data;

[0031] focusing mining the power consumption focusing feature based on the process parameter features to obtain the process power consumption correlation feature of the target workpiece.

[0032] By adopting the above technical solution, in order to obtain the process power consumption correlation feature of the target workpiece, the process parameter data is subjected to semantic embedding to obtain corresponding process parameter embedding features, then the process parameter embedding features are subjected to semantic mining to obtain process parameter features corresponding to the process parameter data, then the power consumption focusing feature is subjected to focusing mining based on the process parameter features to obtain the process power consumption correlation feature of the target workpiece.

[0033] Optionally, the step of performing semantic mining on the process parameter embedding features to obtain process parameter features corresponding to the process parameter data comprises:

[0034] loading the process parameter embedding features into a deep mining unit of the workpiece quality analysis network, wherein the deep mining unit has a first mapping branch and a second mapping branch, and both mapping branches include linear mapping and nonlinear mapping;

[0035] performing self-attention processing on the process parameter embedding features to obtain process parameter attention features corresponding to the process parameter embedding features;

[0036] performing mapping operation on the process parameter attention features through the first mapping branch to obtain first process parameter mapping features corresponding to the process parameter attention features;

[0037] performing mapping operation on the process parameter attention features through the second mapping branch to obtain second process parameter mapping features corresponding to the process parameter attention features;

[0038] performing cross-attention processing on the first process parameter mapping features and the second process parameter mapping features to obtain process parameter features corresponding to the process parameter data.

[0039] By adopting the above technical solution, in order to obtain process parameter features corresponding to process parameter data, process parameter embedding features are loaded into a deep mining unit of a workpiece quality analysis network, wherein the deep mining unit has a first mapping branch and a second mapping branch, and both mapping branches include linear mapping and nonlinear mapping. Then, self-attention processing is performed on the process parameter embedding features to obtain process parameter attention features corresponding to the process parameter embedding features. Then, mapping operation is performed on the process parameter attention features through the first mapping branch to obtain first process parameter mapping features corresponding to the process parameter attention features. Then, mapping operation is performed on the process parameter attention features through the second mapping branch to obtain second process parameter mapping features corresponding to the process parameter attention features. Then, cross-attention processing is performed on the first process parameter mapping features and the second process parameter mapping features to obtain process parameter features corresponding to the process parameter data.

[0040] Optionally, the step of performing focus mining on the power consumption focus features based on the process parameter features to obtain process power consumption correlation features of the target workpiece comprises:

[0041] load the process parameter feature and the power consumption focus feature into a second focus mining unit of the workpiece quality analysis network, wherein the second focus mining unit comprises a first process parameter focus branch, a second process parameter focus branch and a third process parameter focus branch;

[0042] focus mine the power consumption focus feature based on the process parameter feature through the first process parameter focus branch, and output a first process parameter focus feature;

[0043] focus mine the power consumption focus feature based on the process parameter feature through the second process parameter focus branch, and output a second process parameter focus feature;

[0044] focus mine the power consumption focus feature based on the process parameter feature through the third process parameter focus branch, and output a third process parameter focus feature, wherein the first process parameter focus feature, the second process parameter focus feature and the third process parameter focus feature are located in the same semantic space;

[0045] weight and splice the first process parameter focus feature, the second process parameter focus feature and the third process parameter focus feature to obtain a process power consumption correlation feature of the target workpiece.

[0046] By adopting the above technical solution, in order to obtain a process power consumption correlation feature of a target workpiece, the process parameter feature and the power consumption focus feature are loaded into a second focus mining unit of a workpiece quality analysis network, wherein the second focus mining unit comprises a first process parameter focus branch, a second process parameter focus branch and a third process parameter focus branch, then the power consumption focus feature is focus mined based on the process parameter feature through the first process parameter focus branch, and a first process parameter focus feature is output, then the power consumption focus feature is focus mined based on the process parameter feature through the second process parameter focus branch, and a second process parameter focus feature is output, then the power consumption focus feature is focus mined based on the process parameter feature through the third process parameter focus branch, and a third process parameter focus feature is output, wherein the first process parameter focus feature, the second process parameter focus feature and the third process parameter focus feature are located in the same semantic space, and then the first process parameter focus feature, the second process parameter focus feature and the third process parameter focus feature are weight and spliced to obtain a process power consumption correlation feature of the target workpiece.

[0047] Optionally, the step of focus mining the power consumption focus feature based on the process parameter feature through the first process parameter focus branch and outputting the first process parameter focus feature comprises:

[0048] load the process parameter feature and the power consumption focusing feature into the first process parameter focusing branch, wherein the first process parameter focusing branch is built-in with a semantic space conversion matrix;

[0049] convert the process parameter feature through the semantic space conversion matrix to obtain a semantic space conversion feature, wherein the semantic space conversion matrix is used to convert the process parameter feature from a current semantic space to a semantic space where the power consumption focusing feature is located;

[0050] determine an association parameter distribution between the semantic space conversion feature and the power consumption focusing feature, and perform a normalization operation on the association parameter distribution to obtain a normalized distribution;

[0051] perform a weighting processing on the power consumption focusing feature based on the normalized distribution to obtain a first process parameter focusing feature.

[0052] By adopting the above technical solution, in order to obtain the first process parameter focusing feature, the process parameter feature and the power consumption focusing feature are loaded into the first process parameter focusing branch, wherein the first process parameter focusing branch is built-in with a semantic space conversion matrix, then the process parameter feature is converted through the semantic space conversion matrix to obtain a semantic space conversion feature, wherein the semantic space conversion matrix is used to convert the process parameter feature from a current semantic space to a semantic space where the power consumption focusing feature is located, then the association parameter distribution between the semantic space conversion feature and the power consumption focusing feature is determined, and the normalization operation is performed on the association parameter distribution to obtain a normalized distribution, and then the weighting processing is performed on the power consumption focusing feature based on the normalized distribution to obtain the first process parameter focusing feature.

[0053] In a second aspect, the present application also provides an industrial Internet of Things production line data analysis system, which adopts the following technical solution:

[0054] The industrial Internet of Things production line data analysis system comprises a management platform, a sensing network platform and an object platform which are sequentially connected in communication, and the management platform is configured with:

[0055] a data acquisition module configured to acquire process parameter data and equipment power consumption data corresponding to each workpiece processed by a target production line, wherein the process parameter data at least includes temperature time series data and pressure time series data, and the equipment power consumption data is power consumption time series data of a corresponding processing equipment when processing a workpiece;

[0056] a first focusing mining module configured to, for each workpiece, take the workpiece as a target workpiece, perform a first focusing mining on target equipment power consumption data of the target workpiece based on non-target equipment power consumption data corresponding to non-target workpieces, and obtain a power consumption focusing feature of the target workpiece.

[0057] a second focusing mining module, configured to perform second focusing mining on the power consumption focusing feature based on the process parameter data, to obtain a process power consumption correlation feature of the target workpiece;

[0058] a workpiece quality analysis module, configured to analyze corresponding workpiece quality based on the process power consumption correlation feature, to obtain a workpiece quality analysis result.

[0059] In a third aspect, the present application further provides a computer device, which adopts the technical scheme as follows:

[0060] A computer device comprises a memory and a processor, the memory stores a computer program capable of running on the processor, and the processor implements the method in the first aspect when executing the computer program.

[0061] In a fourth aspect, the present application further provides a computer readable storage medium, which adopts the technical scheme as follows:

[0062] A computer readable storage medium stores a computer program capable of being loaded and executed by a processor to implement the method in the first aspect.

[0063] To sum up, the present application at least has the following beneficial technical effects: the process parameter data and the equipment power consumption data corresponding to each workpiece processed by the target production line are obtained, wherein the process parameter data at least includes temperature time sequence data and pressure time sequence data, and the equipment power consumption data is the power consumption time sequence data of the corresponding processing equipment when the workpiece is produced, then for each workpiece, the workpiece is taken as a target workpiece, the target equipment power consumption data of the target workpiece is first focused and mined based on the non-target equipment power consumption data corresponding to the non-target workpiece, to obtain the power consumption focusing feature of the target workpiece, then the power consumption focusing feature is second focused and mined based on the process parameter data, to obtain the process power consumption correlation feature of the target workpiece, and then the corresponding workpiece quality is analyzed based on the process power consumption correlation feature, to obtain the workpiece quality analysis result; the present application effectively overcomes the defects of isolated analysis of workpiece data and insufficient multi-source data fusion in the prior art by introducing a double focusing mining mechanism, so that the quality prediction and defect diagnosis result of the workpiece are more accurate and reliable, and strong decision support is provided for intelligent quality control and process optimization of the production line. BRIEF DESCRIPTION OF DRAWINGS

[0064] Figure 1 is a whole process schematic diagram of an embodiment of the present application.

[0065] Figure 2 is a structure schematic diagram of one of application scenarios of a system of an embodiment of the present application.

[0066] Figure 3 is a structural schematic diagram of another application scenario of the system of the embodiment of the present application.

[0067] Figure 4 is a structural block diagram of the computer device of the present application. DETAILED DESCRIPTION

[0068] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be further described in detail below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0069] The embodiment of the present application discloses an industrial Internet of Things production line data analysis method.

[0070] With reference to Figure 1 , the industrial Internet of Things production line data analysis method is applied to an industrial Internet of Things system, the industrial Internet of Things system includes a management platform, a sensing network platform and an object platform which are sequentially connected in communication, the method is executed by the management platform, and includes the following steps.

[0071] In step S11, process parameter data and equipment power consumption data corresponding to each workpiece of a target production line are acquired.

[0072] The process parameter data at least includes temperature time series data and pressure time series data, and the equipment power consumption data is power consumption time series data of a corresponding processing equipment when a workpiece is produced.

[0073] It should be noted that in step S11, through the Internet of Things sensors deployed on the production line, two categories of time series data corresponding to each workpiece in the processing process are synchronously collected: one is process parameter data reflecting the production environment and state, such as time series of key physical quantities such as temperature, pressure, sound, etc.; the other is equipment power consumption data reflecting the equipment working load and energy efficiency, that is, the sequence of power consumption of a specific equipment changing with time when processing the workpiece, and these data together constitute the data basis for subsequent analysis.

[0074] In step S12, for each workpiece, the workpiece is taken as a target workpiece, and based on non-target equipment power consumption data corresponding to non-target workpieces, first focused mining is performed on target equipment power consumption data of the target workpiece to obtain power consumption focused features of the target workpiece.

[0075] It should be noted that in step S12, the first focus mining refers to placing the power consumption data of one workpiece (target workpiece) in the context of the entire production batch, using the power consumption data of all other workpieces (non-target workpieces) as a reference set, and analyzing the subtle differences and common patterns between the target power consumption and the reference set through advanced models. This process can filter out common power consumption fluctuations, thereby mining and amplifying the unique power consumption characteristics of the current target workpiece that may be related to quality abnormalities, and obtaining more discriminative power consumption focused features.

[0076] Step S13, based on the process parameter data, the power consumption focused features are subjected to second focus mining to obtain process power consumption correlation features of the target workpiece.

[0077] It should be noted that in step S13, the second focus mining refers to correlating the purified power consumption features obtained in the previous step with another dimension of process parameter data (such as temperature, pressure curve), and through the use of focus mining technology (such as cross-attention) again, exploring the complex cross-modal correlation of "how should the power consumption features of the equipment behave under a specific process parameter change pattern", and finally generating a deep feature that can comprehensively reflect the "process-power consumption" collaborative state, i.e. "process power consumption correlation features".

[0078] Step S14, based on the process power consumption correlation features, the corresponding workpiece quality is analyzed to obtain workpiece quality analysis results.

[0079] It should be noted that in step S14, the deep fusion features (process power consumption correlation features) obtained in the previous step are input into a quality analysis model (such as a classifier or regressor). The model learns the complex mapping relationship between this deep feature and the final product quality (such as pass, defect type, performance index) in historical data, accurately predicts, classifies or anomaly detects the quality state of the current workpiece, and outputs specific quality analysis results, thereby completing the transformation from data to knowledge.

[0080] In the above embodiment, the process parameter data and the equipment power consumption data corresponding to each workpiece processed by the target production line are acquired, wherein the process parameter data at least includes temperature time series data and pressure time series data, and the equipment power consumption data is power consumption time series data of a corresponding processing equipment when the workpiece is produced, then for each workpiece, the workpiece is taken as a target workpiece, target equipment power consumption data of the target workpiece is first focused and mined based on non-target equipment power consumption data corresponding to non-target workpieces, power consumption focusing features of the target workpiece are obtained, then the power consumption focusing features are second focused and mined based on the process parameter data, process power consumption correlation features of the target workpiece are obtained, then corresponding workpiece quality is analyzed based on the process power consumption correlation features, and workpiece quality analysis results are obtained; by introducing a double focusing mining mechanism, the defects of isolated analysis of workpiece data and insufficient multi-source data fusion in the prior art are effectively overcome, the quality prediction and defect diagnosis results of the workpiece are more accurate and reliable, and strong decision support is provided for intelligent quality control and process optimization of the production line.

[0081] As a further embodiment of the method, the step of first focusing and mining the target equipment power consumption data of the target workpiece based on the non-target equipment power consumption data corresponding to the non-target workpieces to obtain the power consumption focusing features of the target workpiece comprises:

[0082] In step S21, the target equipment power consumption data is subjected to semantic embedding to obtain corresponding first power consumption embedding features, and the non-target equipment power consumption data corresponding to each non-target workpiece is respectively subjected to semantic embedding to obtain corresponding second power consumption embedding features.

[0083] It should be noted that the original time series data usually has problems such as high dimensionality, redundancy and noise, and the semantic embedding process maps these data into an abstract semantic space. In this space, the distance and relationship between data points can better reflect their inherent and meaningful patterns; in step S21, the equipment power consumption time series data of the target workpiece (the current analysis object) is converted through an embedding layer (such as a linear layer or a neural network) to obtain its corresponding low-dimensional and dense vector representation, i.e. the first power consumption embedding features; at the same time, the same operation is performed on the power consumption data of each non-target workpiece (other reference objects) to generate respective second power consumption embedding features.

[0084] In step S22, the first power consumption embedding features and the second power consumption embedding features are loaded into a first focusing mining unit of the workpiece quality analysis network, wherein the first focusing mining unit includes a first power consumption focusing branch and a second power consumption focusing branch.

[0085] It should be noted that the first power consumption embedding feature (target) obtained in step S21 and all second power consumption embedding features (context background) are input into a special neural network module called a first focus mining unit, which is designed to have two parallel branches: a first power consumption focus branch and a second power consumption focus branch. Through the double-branch parallel structure, different branches focus on extracting and emphasizing different aspects of the target feature from the background information, similar to having multiple experts analyze the same problem from different angles, thereby obtaining a more comprehensive and richer feature representation.

[0086] In step S23, for each second power consumption embedding feature, the first power consumption embedding feature is focused and mined based on the second power consumption embedding feature through the first power consumption focus branch to obtain a first power consumption focus feature. And through the second power consumption focus branch, the first power consumption embedding feature is focused and mined based on the second power consumption embedding feature to obtain a second power consumption focus feature.

[0087] It should be noted that in step S23, the first power consumption embedding feature is subjected to a kind of focus or attention calculation with the second power consumption embedding feature as a reference, outputting a feature reflecting the difference and relevance between the target and the specific non-target workpiece, i.e., the first power consumption focus feature. Similarly, the first power consumption embedding feature is subjected to another focus calculation with the second power consumption embedding feature as a reference, extracting information from another angle to obtain the second power consumption focus feature. By allowing the target data to interact and compare with each background data one by one, the model can finely assess the "normal" and "abnormal" degree of the target workpiece in the overall production background. The two branches work in parallel to ensure that multi-dimensional comparison features are extracted from rich context information, avoiding information loss that may be caused by a single perspective.

[0088] In step S24, the first power consumption focus feature, the second power consumption focus feature, and the first power consumption embedding feature are spliced to obtain the power consumption focus feature of the target workpiece.

[0089] In the above embodiment, in order to obtain the power consumption focusing feature of the target workpiece, the target device power consumption data is subjected to semantic embedding to obtain a corresponding first power consumption embedding feature, and the non-target device power consumption data corresponding to each non-target workpiece is respectively subjected to semantic embedding to obtain a corresponding second power consumption embedding feature, then the first power consumption embedding feature and the second power consumption embedding feature are loaded into a first focusing mining unit of the workpiece quality analysis network, wherein the first focusing mining unit comprises a first power consumption focusing branch and a second power consumption focusing branch, then for each second power consumption embedding feature, through the first power consumption focusing branch, the first power consumption embedding feature is subjected to focusing mining based on the second power consumption embedding feature to obtain a first power consumption focusing feature, and through the second power consumption focusing branch, the first power consumption embedding feature is subjected to focusing mining based on the second power consumption embedding feature to obtain a second power consumption focusing feature, then the first power consumption focusing feature, the second power consumption focusing feature and the first power consumption embedding feature are spliced to obtain the power consumption focusing feature of the target workpiece.

[0090] As a further embodiment of the method, the step of obtaining the first power consumption focusing feature by the first power consumption focusing branch based on the second power consumption embedding feature and the first power consumption embedding feature comprises:

[0091] Step S31, loading the first power consumption embedding feature and the second power consumption embedding feature into the first power consumption focusing branch, wherein the first power consumption focusing branch has a first linear mapping layer, a second linear mapping layer and a third linear mapping layer.

[0092] Step S32, linearly mapping the first power consumption embedding feature based on the first linear mapping layer to generate a query feature vector.

[0093] Step S33, linearly mapping the second power consumption embedding feature through the second linear mapping layer to generate a key feature vector.

[0094] Step S34, linearly mapping the second power consumption embedding feature based on the third linear mapping layer to generate a value feature vector.

[0095] Step S35, determining a similarity matrix between the query feature vector and the key feature vector, and normalizing the similarity matrix to generate a normalized weight matrix.

[0096] Step S36, weighting and summing the value feature vector based on the normalized weight matrix to obtain the first power consumption focusing feature.

[0097] In the above embodiment, in order to obtain the first power consumption focusing feature, the first power consumption embedding feature and the second power consumption embedding feature are loaded into the first power consumption focusing branch, wherein the first power consumption focusing branch has a first linear mapping layer, a second linear mapping layer and a third linear mapping layer, then the first power consumption embedding feature is linearly mapped based on the first linear mapping layer to generate a query feature vector, then the second power consumption embedding feature is linearly mapped through the second linear mapping layer to generate a key feature vector, then the second power consumption embedding feature is linearly mapped based on the third linear mapping layer to generate a value feature vector, then a similarity matrix between the query feature vector and the key feature vector is determined, and the similarity matrix is normalized to generate a normalized weight matrix, and then the value feature vector is weighted and summed based on the normalized weight matrix to obtain the first power consumption focusing feature. In addition, the technical principle of the second power consumption focusing feature is basically the same as that of the first process parameter focusing feature, and the technical principle of the second power consumption focusing feature can refer to steps S71 to S74.

[0098] As a further embodiment of the method, the step of focusing mining the power consumption focusing feature based on the process parameter data to obtain the process power consumption correlation feature of the target workpiece comprises:

[0099] Step S41, performing semantic embedding on the process parameter data to obtain corresponding process parameter embedding features.

[0100] Step S42, performing semantic mining on the process parameter embedding features to obtain process parameter features corresponding to the process parameter data.

[0101] Step S43, focusing mining the power consumption focusing feature based on the process parameter features to obtain the process power consumption correlation feature of the target workpiece.

[0102] In the above embodiment, in order to obtain the process power consumption correlation feature of the target workpiece, the process parameter data is semantically embedded to obtain corresponding process parameter embedding features, then the process parameter embedding features are semantically mined to obtain process parameter features corresponding to the process parameter data, and then the power consumption focusing feature is focused mined based on the process parameter features to obtain the process power consumption correlation feature of the target workpiece.

[0103] As a further embodiment of the method, the step of performing semantic mining on the process parameter embedding features to obtain process parameter features corresponding to the process parameter data comprises:

[0104] Step S51, loading the process parameter embedding features into a deep mining unit of the workpiece quality analysis network, wherein the deep mining unit has a first mapping branch and a second mapping branch, and both mapping branches include linear mapping and nonlinear mapping.

[0105] In step S52, the process parameter embedding features are subjected to self-attention processing to obtain process parameter attention features corresponding to the process parameter embedding features.

[0106] In step S53, the process parameter attention features are subjected to mapping operation through the first mapping branch to obtain first process parameter mapping features corresponding to the process parameter attention features.

[0107] In step S54, the process parameter attention features are subjected to mapping operation through the second mapping branch to obtain second process parameter mapping features corresponding to the process parameter attention features.

[0108] In step S55, the first process parameter mapping features and the second process parameter mapping features are subjected to cross-attention processing to obtain process parameter features corresponding to the process parameter data.

[0109] It should be noted that in steps S51 to S55, the process parameter embedding features are first input into a dedicated deep mining unit, which adopts a double-branch structure to analyze information from different angles; then through self-attention processing, the model first captures the global dependency between different time points in the process parameter sequence, enhances the representation of key time step information, and generates process parameter attention features; then two independent mapping branches respectively perform linear and nonlinear transformation on the attention features; the first branch focuses on mining a feature mode (such as macro trend), and the second branch focuses on capturing another complementary feature mode (such as micro fluctuation), thereby generating two mapping features with different perspectives. Finally, through cross-attention processing, the two feature sequences derived from the same input but focusing on different aspects are deeply interacted, allowing the model to adaptively filter and fuse the most valuable information from the two, thereby outputting a final feature that can comprehensively and deeply represent the process state.

[0110] In the above embodiment, in order to obtain process parameter features corresponding to the process parameter data, the process parameter embedding features are loaded into a deep mining unit of the workpiece quality analysis network, wherein the deep mining unit has a first mapping branch and a second mapping branch, and both mapping branches include linear mapping and nonlinear mapping, then the process parameter embedding features are subjected to self-attention processing to obtain process parameter attention features corresponding to the process parameter embedding features, then the process parameter attention features are subjected to mapping operation through the first mapping branch to obtain first process parameter mapping features corresponding to the process parameter attention features, then the process parameter attention features are subjected to mapping operation through the second mapping branch to obtain second process parameter mapping features corresponding to the process parameter attention features, and then the first process parameter mapping features and the second process parameter mapping features are subjected to cross-attention processing to obtain process parameter features corresponding to the process parameter data.

[0111] As a further implementation form of the method, the step of focusing mining, based on the process parameter feature, on the power consumption focusing feature to obtain the process power consumption correlation feature of the target workpiece comprises:

[0112] In step S61, the process parameter feature and the power consumption focusing feature are loaded into a second focusing mining unit of the workpiece quality analysis network, wherein the second focusing mining unit comprises a first process parameter focusing branch, a second process parameter focusing branch and a third process parameter focusing branch.

[0113] In step S62, the first process parameter focusing feature is output by focusing mining, based on the process parameter feature, on the power consumption focusing feature through the first process parameter focusing branch.

[0114] In step S63, the second process parameter focusing feature is output by focusing mining, based on the process parameter feature, on the power consumption focusing feature through the second process parameter focusing branch.

[0115] In step S64, the third process parameter focusing feature is output by focusing mining, based on the process parameter feature, on the power consumption focusing feature through the third process parameter focusing branch, wherein the first process parameter focusing feature, the second process parameter focusing feature and the third process parameter focusing feature are located in the same semantic space.

[0116] In step S65, the first process parameter focusing feature, the second process parameter focusing feature and the third process parameter focusing feature are weighted and spliced to obtain the process power consumption correlation feature of the target workpiece.

[0117] It should be noted that steps S61 to S65 realize the deep and multi-angle fusion between the process parameter feature and the power consumption focusing feature by designing a second focusing mining unit comprising three parallel branches (first, second and third process parameter focusing branches). The three branches use the same or different internal attention mechanisms to respectively mine the complex correlation between “process-power consumption” from a specific angle and generate three complementary focusing features (first, second and third process parameter focusing features) located in the same semantic space. Finally, an adaptive weighting splicing strategy is used to organically integrate these correlation features mined from different angles to form a final feature that can fully and sufficiently represent the deep coupling relationship between the process state and the equipment power consumption, i.e., the process power consumption correlation feature, thereby enhancing the richness and robustness of multi-modal feature fusion and the accuracy of the analysis result.

[0118] In the above embodiment, in order to obtain the process power consumption correlation feature of the target workpiece, the process parameter feature and the power consumption focusing feature are loaded into the second focusing mining unit of the workpiece quality analysis network, wherein the second focusing mining unit comprises a first process parameter focusing branch, a second process parameter focusing branch and a third process parameter focusing branch, then the power consumption focusing feature is focused and mined based on the process parameter feature through the first process parameter focusing branch, and the first process parameter focusing feature is output, then the power consumption focusing feature is focused and mined based on the process parameter feature through the second process parameter focusing branch, and the second process parameter focusing feature is output, then the power consumption focusing feature is focused and mined based on the process parameter feature through the third process parameter focusing branch, and the third process parameter focusing feature is output, wherein the first process parameter focusing feature, the second process parameter focusing feature and the third process parameter focusing feature are in the same semantic space, then the first process parameter focusing feature, the second process parameter focusing feature and the third process parameter focusing feature are weighted and spliced to obtain the process power consumption correlation feature of the target workpiece.

[0119] As a further embodiment of the method, the step of focusing and mining the power consumption focusing feature based on the process parameter feature through the first process parameter focusing branch to output the first process parameter focusing feature comprises:

[0120] Step S71, loading the process parameter feature and the power consumption focusing feature into the first process parameter focusing branch, wherein the first process parameter focusing branch is built-in with a semantic space conversion matrix.

[0121] Step S72, performing semantic space conversion on the process parameter feature through the semantic space conversion matrix to obtain a semantic space conversion feature, wherein the semantic space conversion matrix is used to convert the process parameter feature from the current semantic space to the semantic space where the power consumption focusing feature is located.

[0122] Step S73, determining the correlation parameter distribution between the semantic space conversion feature and the power consumption focusing feature, and performing a normalization operation on the correlation parameter distribution to obtain a normalized distribution.

[0123] Step S74, performing weighted processing on the power consumption focusing feature based on the normalized distribution to obtain the first process parameter focusing feature.

[0124] It should be noted that steps S71 to S74, by the built-in semantic space conversion matrix, accurately map the process parameter characteristics into the semantic space where the power consumption focusing characteristics are located, realize the alignment of different modal characteristics in the same semantic dimension, then calculate the correlation distribution between the converted process characteristics and the power consumption characteristics, and obtain the attention weight through normalization processing, and finally according to the weight, the power consumption focusing characteristics are finely weighted and fused, to generate the first process parameter focusing characteristics which not only retain the original power consumption characteristics but also deeply fuse the process parameter information. This method effectively solves the fusion difficulty problem caused by inconsistent semantic space of multi-modal characteristics, and realizes accurate cross-modal characteristic interaction. In addition, the technical principles of the second process parameter focusing characteristics and the third process parameter focusing characteristics are basically the same as those of the first process parameter focusing characteristics. The technical principles of the second process parameter focusing characteristics and the third process parameter focusing characteristics can refer to steps S71 to S74.

[0125] In the above embodiment, in order to obtain the first process parameter focusing characteristics, the process parameter characteristics and the power consumption focusing characteristics are loaded into the first process parameter focusing branch, wherein the first process parameter focusing branch is built-in with a semantic space conversion matrix, then the semantic space conversion matrix is used to convert the process parameter characteristics in the semantic space to obtain the semantic space conversion characteristics, wherein the semantic space conversion matrix is used to convert the process parameter characteristics from the current semantic space to the semantic space where the power consumption focusing characteristics are located, then the correlation parameter distribution between the semantic space conversion characteristics and the power consumption focusing characteristics is determined, and the correlation parameter distribution is normalized to obtain a normalized distribution, and then the power consumption focusing characteristics are weighted based on the normalized distribution to obtain the first process parameter focusing characteristics.

[0126] The application also discloses an industrial Internet of Things production line data analysis system.

[0127] Reference Figure 2 The industrial Internet of Things production line data analysis system comprises a management platform, a sensing network platform and an object platform which are sequentially connected in communication, and the management platform is configured to have:

[0128] A data acquisition module is configured to acquire process parameter data and equipment power consumption data corresponding to each workpiece processed by a target production line, wherein the process parameter data at least comprises temperature time series data and pressure time series data, and the equipment power consumption data is power consumption time series data of corresponding processing equipment when the workpiece is processed;

[0129] A first focusing mining module is configured to, for each workpiece, take the workpiece as a target workpiece, perform first focusing mining on target equipment power consumption data of the target workpiece based on non-target equipment power consumption data corresponding to non-target workpieces, and obtain power consumption focusing characteristics of the target workpiece;

[0130] a second focus mining module configured to perform second focus mining on the power consumption focus feature based on the process parameter data to obtain a process power consumption correlation feature of the target workpiece;

[0131] a workpiece quality analysis module configured to analyze the corresponding workpiece quality based on the process power consumption correlation feature to obtain a workpiece quality analysis result.

[0132] Another application scenario of the industrial Internet of Things production line data analysis system is shown in FIG. 6, which can include a user platform, a service platform, a management platform, a sensing network platform, and an object platform that interact in sequence to form a five-platform architecture based on the industrial Internet of Things. Figure 3 The management platform includes a data acquisition module, a first focus mining module, a second focus mining module, and a workpiece quality analysis module. The sensing network platform includes a sensing total database and a plurality of sensing network sub-platforms. Each sensing network sub-platform can communicate with the sensing total database, and each sensing network sub-platform is provided with a sensing sub-database.

[0133] Specifically, in the above-mentioned another application scenario, the industrial Internet of Things production line data analysis system includes a management platform, which is configured to acquire process parameter data and equipment power consumption data corresponding to each workpiece processed by a target production line, wherein the process parameter data at least includes temperature time series data and pressure time series data, and the equipment power consumption data is power consumption time series data of a corresponding processing equipment when processing a workpiece. For each workpiece, the workpiece is taken as a target workpiece, the target equipment power consumption data of the target workpiece is subjected to first focus mining based on non-target equipment power consumption data corresponding to a non-target workpiece, to obtain a power consumption focus feature of the target workpiece. The power consumption focus feature is subjected to second focus mining based on the process parameter data, to obtain a process power consumption correlation feature of the target workpiece. The corresponding workpiece quality is analyzed based on the process power consumption correlation feature, to obtain a workpiece quality analysis result.

[0134] Through the interaction between the various functional platforms of the three-platform or five-platform based industrial Internet of Things production line data analysis system, a perfect closed-loop information operation logic is established to ensure the orderly operation of the sensing information and the control information, and to realize intelligent management of the equipment.

[0135] The industrial Internet of Things production line data analysis system of the present application can implement any of the methods in the industrial Internet of Things production line data analysis method, and the specific working process of the industrial Internet of Things production line data analysis system of the present application can refer to the corresponding process in the above-mentioned industrial Internet of Things production line data analysis method.

[0136] The present application also discloses a computer device.

[0137] Reference is made to Figure 4A computer device includes a memory and a processor, the memory has a computer program stored thereon, the computer program is capable of being run on the processor, and the processor implements any one of the industrial internet of things production line data analysis methods when executing the computer program.

[0138] The embodiments of the present application further disclose a computer readable storage medium.

[0139] A computer readable storage medium stores a computer program capable of being loaded and executed by a processor to execute any one of the industrial internet of things production line data analysis methods.

[0140] The computer readable storage medium can be any tangible medium that contains or stores a program for use by or in connection with an instruction execution system, apparatus, or device; and the program code contained in the computer readable medium can be transmitted in any appropriate medium, including but not limited to wireless, wire, optical cable, RF, etc., or any appropriate combination of the above.

[0141] The above are preferred embodiments of the present application, and are not intended to limit the protection scope of the present application; any feature disclosed in the specification (including the abstract and drawings) can be replaced by other equivalent or similar features unless specifically described, and each feature is only an example of a series of equivalent or similar features. That is, unless specifically described, each feature is only an example of a series of equivalent or similar features.

Claims

1. A data analysis method for industrial IoT production lines, characterized in that, Applied to an industrial Internet of Things (IIoT) system, the IIoT system includes a management platform, a sensor network platform, and an object platform that are sequentially connected in communication. The method is executed by the management platform and includes: Obtain the process parameter data and equipment power consumption data corresponding to each workpiece processed by the target production line. The process parameter data includes at least temperature time series data and pressure time series data, and the equipment power consumption data is the power consumption time series data of the corresponding processing equipment when producing the workpiece. For each workpiece, the workpiece is taken as the target workpiece. Based on the power consumption data of the non-target devices corresponding to the non-target workpieces, the power consumption data of the target devices of the target workpieces is subjected to the first focusing mining to obtain the power consumption focusing features of the target workpieces. Based on the process parameter data, a second focusing mining is performed on the power consumption focusing features to obtain the process power consumption correlation features of the target workpiece. Based on the process power consumption correlation characteristics, the corresponding workpiece quality is analyzed to obtain the workpiece quality analysis results; The step of performing a first focusing mining on the power consumption data of the target device of the target workpiece based on the power consumption data of the non-target device corresponding to the non-target workpiece to obtain the power consumption focusing features of the target workpiece includes: Semantic embedding is performed on the power consumption data of the target device to obtain the corresponding first power consumption embedding feature, and semantic embedding is performed on the power consumption data of the non-target device corresponding to each non-target workpiece to obtain the corresponding second power consumption embedding feature. The first power consumption embedding feature and the second power consumption embedding feature are loaded into the first focusing mining unit of the workpiece quality analysis network, wherein the first focusing mining unit includes a first power consumption focusing branch and a second power consumption focusing branch; For each second power consumption embedding feature, the first power consumption embedding feature is focused and mined based on the second power consumption embedding feature through the first power consumption focusing branch to obtain the first power consumption focusing feature; and the second power consumption focusing feature is focused and mined based on the second power consumption embedding feature through the second power consumption focusing branch to obtain the second power consumption focusing feature. The power consumption focusing feature, the second power consumption focusing feature, and the first power consumption embedding feature are spliced ​​together to obtain the power consumption focusing feature of the target workpiece.

2. The industrial IoT production line data analysis method according to claim 1, characterized in that, The step of obtaining the first power consumption focusing feature by performing a first focusing mining on the first power consumption embedding feature based on the second power consumption embedding feature through the first power consumption focusing branch includes: The first power consumption embedding feature and the second power consumption embedding feature are loaded into the first power consumption focusing branch, wherein the first power consumption focusing branch has a first linear mapping layer, a second linear mapping layer and a third linear mapping layer; Based on the first linear mapping layer, the first power consumption embedding feature is linearly mapped to generate a query feature vector; The second linear mapping layer performs a linear mapping on the second power consumption embedding feature to generate a key feature vector; Based on the third linear mapping layer, the second power consumption embedding feature is linearly mapped to generate a value feature vector; Determine the similarity matrix between the query feature vector and the key feature vector, and normalize the similarity matrix to generate a normalized weight matrix; Based on the normalized weight matrix, the value feature vector is weighted and summed to obtain the first power consumption focusing feature.

3. The industrial IoT production line data analysis method according to claim 1, characterized in that, The step of focusing and mining the power consumption focusing features based on the process parameter data to obtain the process power consumption correlation features of the target workpiece includes: Semantic embedding is performed on the process parameter data to obtain the corresponding process parameter embedding features; Semantic mining is performed on the embedded features of the process parameters to obtain the process parameter features corresponding to the process parameter data. Based on the process parameter characteristics, the power consumption focusing characteristics are focused and mined to obtain the process power consumption correlation characteristics of the target workpiece.

4. The industrial IoT production line data analysis method according to claim 3, characterized in that, The step of performing semantic mining on the embedded features of the process parameters to obtain the process parameter features corresponding to the process parameter data includes: The process parameters are embedded features and loaded into the deep mining unit of the workpiece quality analysis network, wherein the deep mining unit has a first mapping branch and a second mapping branch, and both mapping branches include linear mapping and nonlinear mapping. Self-attention processing is performed on the process parameter embedding features to obtain the process parameter attention features corresponding to the process parameter embedding features; The process parameter attention features are mapped through the first mapping branch to obtain the first process parameter mapping features corresponding to the process parameter attention features. The process parameter attention features are mapped using the second mapping branch to obtain the second process parameter mapping features corresponding to the process parameter attention features. Cross-attention processing is performed on the first process parameter mapping feature and the second process parameter mapping feature to obtain the process parameter feature corresponding to the process parameter data.

5. The industrial IoT production line data analysis method according to claim 3, characterized in that, The step of focusing and mining the power consumption focusing features based on the process parameter features to obtain the process power consumption correlation features of the target workpiece includes: The process parameter features and the power consumption focusing features are loaded into the second focusing mining unit of the workpiece quality analysis network, wherein the second focusing mining unit includes a first process parameter focusing branch, a second process parameter focusing branch and a third process parameter focusing branch; By focusing on the first process parameter branch, focusing on the power consumption focusing feature based on the process parameter features, and outputting the first process parameter focusing feature; By using the second process parameter focusing branch, the power consumption focusing feature is mined based on the process parameter features, and the second process parameter focusing feature is output. By using the third process parameter focusing branch, the power consumption focusing feature is focused and mined based on the process parameter features, and the third process parameter focusing feature is output. The first process parameter focusing feature, the second process parameter focusing feature and the third process parameter focusing feature are located in the same semantic space. The first process parameter focusing feature, the second process parameter focusing feature, and the third process parameter focusing feature are weighted and spliced ​​to obtain the process power consumption correlation feature of the target workpiece.

6. The industrial IoT production line data analysis method according to claim 5, characterized in that, The step of focusing on the power consumption focusing features based on the process parameter features through the first process parameter focusing branch and outputting the first process parameter focusing features includes: The process parameter features and the power consumption focusing features are loaded into the first process parameter focusing branch, wherein the first process parameter focusing branch has a built-in semantic space transformation matrix; The semantic space transformation matrix is ​​used to perform semantic space transformation on the process parameter features to obtain semantic space transformation features. The semantic space transformation matrix is ​​used to transform the process parameter features from the current semantic space to the semantic space where the power consumption focusing features are located. Determine the distribution of correlation parameters between the semantic space transformation features and the power consumption focusing features, and normalize the distribution of correlation parameters to obtain a normalized distribution; Based on the normalized distribution, the power consumption focusing features are weighted to obtain the first process parameter focusing features.

7. An industrial IoT production line data analysis system, characterized in that, It includes a management platform, a sensor network platform, and an object platform that are connected in sequence. The management platform is configured with: The data acquisition module is used to acquire the process parameter data and equipment power consumption data corresponding to each workpiece processed by the target production line. The process parameter data includes at least temperature time series data and pressure time series data, and the equipment power consumption data is the power consumption time series data of the corresponding processing equipment when producing the workpiece. The first focused mining module is used to, for each workpiece, take the workpiece as the target workpiece, and perform first focused mining on the target device power consumption data of the target workpiece based on the non-target device power consumption data corresponding to the non-target workpiece, so as to obtain the power consumption focused features of the target workpiece. The second focus mining module is used to perform second focus mining on the power consumption focus features based on the process parameter data to obtain the process power consumption correlation features of the target workpiece. The workpiece quality analysis module is used to analyze the corresponding workpiece quality based on the process power consumption correlation characteristics and obtain the workpiece quality analysis results. The step of performing a first focusing mining on the power consumption data of the target device of the target workpiece based on the power consumption data of the non-target device corresponding to the non-target workpiece to obtain the power consumption focusing features of the target workpiece includes: Semantic embedding is performed on the power consumption data of the target device to obtain the corresponding first power consumption embedding feature, and semantic embedding is performed on the power consumption data of the non-target device corresponding to each non-target workpiece to obtain the corresponding second power consumption embedding feature. The first power consumption embedding feature and the second power consumption embedding feature are loaded into the first focusing mining unit of the workpiece quality analysis network, wherein the first focusing mining unit includes a first power consumption focusing branch and a second power consumption focusing branch; For each second power consumption embedding feature, the first power consumption embedding feature is focused and mined based on the second power consumption embedding feature through the first power consumption focusing branch to obtain the first power consumption focusing feature; and the second power consumption focusing feature is focused and mined based on the second power consumption embedding feature through the second power consumption focusing branch to obtain the second power consumption focusing feature. The power consumption focusing feature, the second power consumption focusing feature, and the first power consumption embedding feature are spliced ​​together to obtain the power consumption focusing feature of the target workpiece.

8. A computer device, characterized in that, The method includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the method of any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer program is stored that can be loaded by a processor and execute the method of any one of claims 1 to 6.

Images