Manufacturing process-oriented large-scale data mining method and system and storage medium

Abstract

The invention discloses a manufacturing process-oriented large-scale data mining method and system and a storage medium, adopts a neural network algorithm based on gradient accumulation compensation, can effectively avoid the problems of gradient explosion and gradient disappearance encountered by a traditional neural network in large-scale process data training, and can improve the data mining efficiency through gradient accumulation and dynamic compensation. The model can stably update parameters, accelerates the convergence process of complex process data, and has higher adaptability and robustness.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent manufacturing, and in particular relates to a large-scale data mining method, system and storage medium for manufacturing processes. Background Art

[0002] With the increasing intelligence of industrial equipment and complexity of production processes, traditional machine learning models are often unable to effectively cope with the challenges encountered in actual production when processing multidimensional data, dynamically changing data, and noisy data, resulting in insufficient generalization of the model, which in turn affects equipment maintenance, fault diagnosis, and production efficiency. Currently, data in many industrial applications usually involves a large number of feature points and has problems such as high dimensionality, redundancy, and noise.

[0003] The Chinese invention patent with the prior art publication number CN119089141A proposes a method and system for optimizing water treatment based on machine learning. The method first collects process variable data such as water quality, flow, liquid level, and pressure through sensors arranged at different parts of the water treatment equipment; the collected process variable data are stored according to type and structure storage to realize data partitioning and indexing; then the stored data is cleaned and standardized, a CNN-LSTM algorithm model is constructed, and the data is stream processed through the CNN-LSTM algorithm model; finally, the results predicted by the model are compared with the preset rules, and the control strategy is confirmed according to the comparison results. According to the confirmed control strategy, the water treatment equipment in each process link is controlled to achieve the overall water treatment effect. The method realizes the accurate prediction, real-time monitoring and scheduling of water treatment process variables, realizes the optimization goals, energy efficiency, and automatic operation and unattended operation of the water treatment system, and significantly improves the level of intelligence, processing efficiency and response speed.

[0004] However, the training process may encounter difficulties in convergence due to abnormal changes in gradients, and may even fail to effectively capture key information in the data. Especially in the processing of complex and multi-dimensional industrial data, existing methods have low stability and efficiency when processing high-dimensional data.

[0005] At the same time, the problem of sample imbalance is ignored, resulting in minority samples being often ignored or not fully learned, which may affect the accuracy of fault diagnosis and further affect production efficiency and equipment operation and maintenance level. Summary of the invention

[0006] The present invention proposes a large-scale data mining method for manufacturing processes, which realizes parameter update optimization by means of gradient accumulation compensation, thereby overcoming the gradient explosion and gradient vanishing problems that are prone to occur in traditional neural networks in large-scale process data training.

[0007] In order to solve the above problems, the technical solution provided by the present invention is: In a first aspect, a large-scale data mining method for manufacturing processes includes data collection, wherein the data is data obtained in the manufacturing process; Extracting features from the data, wherein a neural network algorithm based on gradient accumulation compensation is used as a feature extraction model; the neural network based on gradient accumulation compensation is obtained through training; Classifying the data according to the features, wherein a classification model is established using a machine learning algorithm to identify the data; Storing and managing the classified data; The data are visualized.

[0008] Furthermore, the training method of the neural network algorithm based on gradient accumulation compensation includes: Initialization of neural network parameters, wherein the parameters of the neural network are initialized by random normal distribution and the initial value of the gradient accumulation is set to zero; avoiding the instability of the model due to premature convergence or oscillation in the initial state, In the back propagation process of each training, the gradient accumulation strategy is adopted to add and accumulate the current gradient and the historical gradient, and smoothly update it through the dynamic compensation factor; the calculation method of the gradient accumulation strategy is expressed as:

[0009]

[0010] In the formula, For the The gradient accumulation of iterations, For the The gradient accumulation of iterations, is a positive integer, is the gradient accumulation attenuation factor, The loss function of the neural network is parameterized The gradient of For the training samples; For the The labels of the training samples; is the loss function of the neural network; For the The local gradient offset for the iteration; are neural network parameters, including weights and bias parameters of the neural network; through gradient accumulation compensation, the model can better capture these small changes and improve the ability to identify the equipment status. By accumulating past data gradients, the model can be more sensitive to the initial abnormal signals in the production process;

[0011] Local gradient offset calculation, wherein the local gradient offset calculation performs weighted processing on important samples through sample weighting coefficients; Use the neural network's loss function to update the neural network parameter gradient; During the feedforward calculation of the neural network, the activation function is dynamically adjusted, and the activation output is combined with the current accumulated gradient information and the scale of the input sample features; Dynamically adjust the learning rate according to the gradient changes during training; Perform global updates on neural network parameters; The above steps are iterated repeatedly, and the training is stopped if the continuous change amplitude of the loss function is less than a given threshold, thereby completing the neural network training based on gradient accumulation compensation.

[0012] Furthermore, the activation function is dynamically adjusted and calculated as follows:

[0013]

[0014] In the formula, For the The activation value of a sample in the neural network, is a positive integer, is the Sigmoid activation function, is the number of neurons in the neural network, is the first forward propagation hyperparameter, is the second forward propagation hyperparameter, For the The cumulative gradient of the iterations, For the training samples, is the norm of the global features in the training dataset, is the L2 norm, The neural network weights, is a positive integer, The neural network A bias.

[0015] Furthermore, an adaptive learning rate adjustment strategy based on batch gradient norm statistics dynamically controls the adaptive learning rate; the adaptive learning rate adjustment strategy based on batch gradient norm statistics is calculated as follows:

[0016]

[0017] In the formula, For the neural network The learning rate of the iteration, is the initial learning rate of the neural network, is the number of samples input to the neural network in the current batch, is a positive integer, is the learning rate adjustment factor, is the threshold for learning rate adjustment, For the The gradient accumulation for iterations.

[0018] Furthermore, the update method of the loss function of the neural network to the parameter gradient is expressed as:

[0019] In the formula, is the parameter update operation, is the gradient update adjustment factor, is the preset gradient threshold, is a hyperparameter that controls the nonlinear scaling of the gradient.

[0020] Furthermore, the classification model structure is based on a support vector machine algorithm.

[0021] Furthermore, it is characterized in that the data obtained in the manufacturing process includes production equipment sensor data, production process monitoring system data, equipment maintenance system data, quality inspection equipment data, and manually recorded data.

[0022] Secondly, a large-scale data mining method for manufacturing process is developed into a large-scale data mining system for manufacturing process, including A data acquisition module, wherein the data acquisition module is used to acquire data in the manufacturing process; A data processing module, the data processing module is used to extract features from the data; A data classification module, wherein the data classification module is used to classify the data according to the features; Data storage and management module: The data storage and management module is used to store and manage the classified data; A data visualization and analysis module is used to process and output the data visualization.

[0023] In a third aspect, a computer-readable storage medium storing instructions is provided, wherein the storage medium stores a computer program or instructions, and when the computer program or instructions are executed by an image processing device, the above-mentioned large-scale data mining method for manufacturing processes is implemented.

[0024] In a fourth aspect, a computer program product is provided, characterized in that the computer program product comprises: a computer program code, which, when executed, enables a processor to execute the above-mentioned large-scale data mining method for manufacturing processes.

[0025] Beneficial effects of the present invention:

[0026] 1. The present invention adopts a neural network algorithm based on gradient accumulation compensation as a feature extraction model, and realizes parameter update optimization by means of gradient accumulation compensation, thereby overcoming the gradient explosion and gradient vanishing problems that are prone to occur in traditional neural networks in large-scale process data training, thereby accelerating the convergence process of complex process data and showing stronger adaptability and robustness.

[0027] 2. Through gradient accumulation compensation, the model can better capture these tiny changes and improve its ability to identify equipment status. By accumulating past data gradients, the model can be more sensitive to initial signals of abnormalities in the production process.

[0028] 3. Adjust the current gradient through sample importance. The calculation of local gradient offset uses sample weighting coefficient to weight important samples, so as to better capture the key features related to the equipment maintenance system. The data in the equipment maintenance system often includes some rare fault information. The samples of this information are few but extremely important. By dynamically adjusting the sample weights, the model can improve the recognition rate of these rare events, thereby optimizing the fault diagnosis capability of the equipment.

[0029] 4. Through the implicit gradient truncation mechanism, excessive or unstable gradients will be truncated, thereby ensuring stable training of neural networks in high-dimensional and nonlinear data. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 Comparison chart of training loss for different optimization methods; Figure 2 This is a comparison chart of gradient stability; Figure 3 This is a comparison chart of the accuracy of mainstream models such as ResNet and LSTM on industrial sensor datasets; Figure 4 Comparison chart of classification accuracy under different noise conditions. DETAILED DESCRIPTION

[0031] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0032] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0033] The present invention proposes a large-scale data mining method for manufacturing processes, which mainly includes the following modules:

[0034] (1) Data acquisition module

[0035] The function of the data acquisition module is to obtain relevant raw data from the manufacturing process. The data source in the present invention is mainly the industrial data generated in the manufacturing process. Various types of machines, equipment, sensors and various links of the production line in the manufacturing process will generate a large amount of data. These data contain information about equipment status, environmental parameters, production efficiency, etc. The data collection sources include: 1) Production equipment sensors: During operation, the equipment uses sensors to record physical quantities such as temperature, pressure, speed, and vibration; 2) Production process monitoring system: real-time monitoring data records key nodes in the production process, including operating parameters, raw material usage, production progress, etc.; 3) Equipment maintenance system: records equipment maintenance history, fault logs, repair time and other data; 4) Quality testing equipment: test data generated by automated testing equipment, such as product size, surface quality, etc.; 5) Manually recorded data: Data recorded through manual inspections can supplement data that cannot be covered by sensors or when abnormalities occur.

[0036] The data collection method is to automatically collect raw data from various sensors and control systems in real time through the combination of industrial control systems and Internet of Things devices, and transmit it to the data storage platform wirelessly or wiredly.

[0037] The collected data is saved in a distributed database in JSON format. Each data record contains multi-dimensional information such as device ID, timestamp, operating parameters, status information, etc., and is structured and stored in a predefined standard format.

[0038] In order to enable the data mining model to achieve supervised training, the collected data is labeled manually. The labeling categories include: 1) Category 1: Abnormal state, equipment failure, process deviation, environmental mutation, etc. 2) Category 2: Normal state, the process is in normal state.

[0039] (2) Feature Engineering Module The function of the feature engineering module is to use neural network technology to extract effective features from data and improve the performance of machine learning models.

[0040] The present invention adopts a neural network algorithm based on gradient accumulation compensation as a feature extraction model, and realizes parameter update optimization by means of gradient accumulation compensation, thereby overcoming the gradient explosion and gradient vanishing problems that are prone to occur in traditional neural networks in large-scale process data training, thereby accelerating the convergence process of complex process data and showing stronger adaptability and robustness.

[0041] Specifically, the training process of the neural network algorithm based on gradient accumulation compensation is as follows: 1. Initialize the parameters of the neural network by random normal distribution, and set the initial value of the gradient accumulation to zero to avoid the model from converging prematurely or oscillating instability in the initial state. For example, in the process data such as temperature and pressure collected by the production equipment sensor, due to the large fluctuation of the initial data, the traditional model may be overfitted or unstable due to the fluctuation of the process data. The initialization strategy of the present invention can effectively avoid this problem. The initialization method is expressed as:

[0042]

[0043]

[0044] In the formula, The neural network weights, is a positive integer, It means the mean is 0 and the variance is The normal distribution of The neural network A bias, is the initialization variance of the neural network, Denotes that it obeys a specific distribution. Preferably, Set to 0.1.

[0045] Furthermore, in the back propagation process of each training, the gradient accumulation strategy is adopted to add and accumulate the current gradient and the historical gradient, and the dynamic compensation factor is used to smoothly update and reduce the instability caused by the sharp fluctuation of the gradient of a single sample. For example, for the time series data in the production process monitoring system, the monitoring system may have instantaneous data fluctuations. Through gradient accumulation compensation, the model can better capture these tiny changes and improve the ability to identify the equipment status. By accumulating past data gradients, the model can be made more sensitive to the initial abnormal signals in the production process. The calculation method is expressed as:

[0046]

[0047] In the formula, For the The gradient accumulation of iterations, For the The gradient accumulation of iterations, is a positive integer, is the gradient accumulation attenuation factor, The loss function of the neural network is parameterized The gradient of For the The labels of the training samples; For the The labels of the training samples; is the loss function of the neural network; For the The local gradient offset for the iteration; are neural network parameters, including weights and bias parameters of the neural network. Preferably, Set to 0.95.

[0048] It should be noted that conventional technologies do not adopt gradient accumulation strategies. When processing large-scale process data, they often encounter gradient vanishing or gradient exploding problems, which will cause the model to converge slowly or even fail to converge during training. In process data feature extraction tasks, such as the predictive model of equipment maintenance systems, due to the presence of a large number of nonlinear features in process data, traditional methods are difficult to effectively solve these problems, resulting in low feature extraction efficiency and even failure to effectively capture the key status of the equipment.

[0049] Furthermore, the current gradient is adjusted by the sample importance, and the calculation of the local gradient offset is to weight the important samples through the sample weighting coefficient, so as to better capture the key features related to the equipment maintenance system. The data in the equipment maintenance system often includes some rare fault information. The samples of this information are few but extremely important. By dynamically adjusting the sample weights, the model can improve the recognition rate of these rare events, thereby optimizing the fault diagnosis ability of the equipment. For example, some extreme fault modes may appear less frequently in historical data, and traditional models often ignore these data. After adopting the weighting strategy, the model can pay more attention to these "rare but important" fault modes during the learning process. The calculation method of the local gradient offset is expressed as:

[0050]

[0051] In the formula, is the sample weighting coefficient, is the L2 norm, is the norm of the current gradient, is the regularization gain factor; For the The feature mean of training samples; is the standard deviation hyperparameter of the sample. Preferably, Set to 0.2, Set to 0.1, Set to 0.2.

[0052] Furthermore, an implicit gradient truncation mechanism is used to judge and truncate excessive or unstable gradients to ensure the stability of high-dimensional process data training and prevent gradient expansion or excessive shrinkage in complex models. The process data generated by quality inspection equipment is often very high-dimensional and noisy. Traditional neural networks may be unstable in training due to the interference of some abnormal process data points. Through the implicit gradient truncation mechanism, excessive or unstable gradients will be truncated, thereby ensuring the stable training of neural networks in high-dimensional and nonlinear data. For example, when modeling the quality inspection data of chemical products, some inspection equipment may cause data anomalies due to improper operation or external environmental influences. This mechanism can effectively avoid the interference of these abnormal process data on model training. The update method of the parameter gradient of the neural network loss function is expressed as:

[0053]

[0054] In the formula, is the parameter update operation, is the gradient update adjustment factor, is the preset gradient threshold, is a hyperparameter that controls the nonlinear scaling of the gradient. Preferably, Set to 0.5, Set to 2, Set to 0.1.

[0055] 2. During the feedforward calculation of the neural network, the activation function is dynamically adjusted so that the activation output is combined with the current gradient accumulation information and the scale of the input sample features to achieve more flexible mapping of high-dimensional and nonlinear process data. Production process monitoring data usually has strong nonlinear characteristics, especially under the influence of multivariate cross-influence. Traditional methods may not be able to fully capture this complexity. By combining the current gradient information and input features, the model can adapt to complex data more flexibly and optimize the mapping effect of the data. For example, in the production monitoring of equipment, the change of equipment status is often affected by multiple factors. By dynamically adjusting the activation function, the complex relationship between these multiple factors can be better revealed. The calculation method is expressed as:

[0056]

[0057] In the formula, For the The activation value of a sample in the neural network, is a positive integer, is the Sigmoid activation function, is the number of neurons in the neural network, is the first forward propagation hyperparameter, is the second forward propagation hyperparameter, For the The gradient accumulation of iterations, For the training samples, is the norm of the global features in the training dataset, is the L2 norm. Preferably, Set to 0.1, Set to 0.3.

[0058] It should be noted that traditional neural network models usually use fixed activation functions when processing high-dimensional and nonlinear data. easy As a result, when faced with complex process data, the model’s mapping capability is insufficient and it is unable to effectively capture the nonlinear features in the data. Especially in quality inspection equipment, there may be complex nonlinear relationships in the data, and fixed activation functions cannot flexibly adapt to high-dimensional and nonlinear data, resulting in the model being unable to effectively map the complex features in the production process, limiting the model’s performance.

[0059] 3. Adopt an adaptive learning rate adjustment strategy based on batch gradient norm statistics, dynamically adjust the learning rate according to the gradient changes during the training process, and ensure faster and more stable convergence speed in complex scenarios. Due to the real-time and variability of monitoring data in the production process, the traditional fixed learning rate method may not perform well in complex scenarios. Dynamically adjusting the learning rate can speed up the convergence speed while maintaining the stability of training. For the equipment maintenance system, dynamic adjustment during the training process can ensure that the model can quickly adjust the learning strategy when facing sudden abnormal equipment process data. The calculation method is expressed as:

[0060]

[0061] In the formula, For the neural network The learning rate of the iteration, is the initial learning rate of the neural network, is the number of samples input to the neural network in the current batch, is a positive integer, is the learning rate adjustment factor, is the threshold for learning rate adjustment, For the Iteration gradient accumulation. Preferably, Set to 0.001, Set to 2, Set to 0.95.

[0062] 4. Globally update the network parameters to converge to a better solution globally. The calculation method is expressed as:

[0063]

[0064] In the formula, For the neural network The parameters of the iterations, For the neural network The parameters of the iterations, For the neural network The learning rate of the iteration, For the The gradient accumulation for iterations.

[0065] 5. Repeat the above steps and determine whether the training can be terminated based on the stability of the training error. If the continuous change of the loss function is less than the given threshold, the model is considered to have reached convergence and training is stopped. The calculation method is expressed as:

[0066]

[0067] In the formula, is the preset minimum allowable error change, For the training samples, For the The labels of the training samples; For the The labels of the training samples; For the The labels of the training samples; is the loss function of the neural network; preferably, Set to 0.001.

[0068] Furthermore, in order to verify the effectiveness of this method, the experimental data are analyzed as follows Figure 1 As shown in the figure, the training loss comparison chart shows that this method has a faster convergence speed than optimization methods such as Adam, SGD, and RMSprop. In addition, the slope of the convergence curve of this method is larger and it enters the stable stage earlier.

[0069] In addition, as follows Figure 2 As shown in the figure, the gradient stability analysis diagram shows that the gradient norm of this method is always stable in a reasonable range, and the convergence process is more stable.

[0070] In addition, as follows Figure 3 As shown in the figure, compared with mainstream models such as ResNet and LSTM on the industrial sensor dataset, this method has the highest accuracy (94.3%) and the smallest standard deviation (±1.2%), and has higher robustness.

[0071] In addition, as follows Figure 4 As shown in the figure, the noise robustness analysis shows that as the input noise increases, the accuracy of this method decreases more slowly, and it still maintains an accuracy of more than 75% under 0.5 intensity noise, which is better than the traditional method.

[0072] (3) Machine Learning Modeling Module

[0073] The role of the machine learning modeling module is to perform classification tasks based on data, use machine learning algorithms to build classification models, and help identify different categories and patterns in the data.

[0074] (4) Data storage and management module

[0075] The function of the data storage and management module is to efficiently store and manage the large amount of data generated by the system, ensuring the security, accessibility and scalability of the data.

[0076] The functions of the data storage and management module include: efficient data storage, supporting distributed storage and fast reading of large-scale data; data backup and recovery, ensuring that data will not be lost in unexpected situations and can be quickly restored; data permission management, ensuring that data access and operation permissions meet security standards and prevent illegal access; data indexing and retrieval, improving data query efficiency through indexing technology to meet the needs of fast data retrieval.

[0077] The machine learning classifier model structure adopted by the present invention is based on the support vector machine algorithm, and the training process is as follows:

[0078] 1. In the training process of the support vector machine, the redundant features in the training data must be processed first. The present invention adopts a redundant elimination mechanism based on information entropy. By calculating the correlation and redundancy between features, the feature space of the data set is dynamically adjusted. Process data usually contains high-dimensional and interrelated features, such as sensor data such as temperature, pressure, and vibration of equipment. If directly input into the support vector machine model, it will bring redundancy, thereby affecting the performance of the model. The redundant elimination loss function can automatically identify and delete redundant features, retain the features that have the greatest impact on the classification results, reduce the computational complexity, and improve the classification accuracy. The calculation method of the redundant elimination loss function is expressed as:

[0079]

[0080] In the formula, Eliminate loss function for redundancy; Features and The mutual information between them measures the correlation between the two; Features and The redundancy coefficient is adaptively adjusted according to its importance; and They represent the first and Features.

[0081] Furthermore, the calculation method of the redundancy coefficient is expressed as:

[0082]

[0083] In the formula, Features and Redundancy measurement between ; Features and Redundancy measurement between ; represents the first Features.

[0084] Furthermore, the calculation method of the redundancy measure is expressed as:

[0085]

[0086]

[0087] In the formula, represents the L2 norm; is the maximum value function.

[0088] 2. In order to deal with the problem of category imbalance, the present invention adopts a dynamic balancing strategy, so that the category ratio of training data is adaptively adjusted as the training process changes. and categories In the current dataset, there are and samples, categories Significantly less than category , then the category weight and Adjusted to:

[0089]

[0090]

[0091] In the formula, and Respectively, categories and The weight of is the number of samples of the category with more samples in the category; and Respectively, categories and categories The number of samples.

[0092] Furthermore, during the training process, the samples with larger category weights are fed earlier into the support vector machine batch. The dynamic balancing strategy can effectively prevent the model from being biased towards categories with larger numbers of samples and improve the classification accuracy of minority classes.

[0093] 3. In the training process of the support vector machine, the selection of samples as support vectors is crucial to the model performance. The traditional support vector machine uses the global optimal support vector. The present invention dynamically selects the most representative support vector through the regularization term. The optimization objective function is:

[0094]

[0095] In the formula, is the loss function of the support vector machine; is the number of samples in the training set; For the The labels of the samples; For the The feature vector of samples; is the weight of the support vector machine model; is the bias of the support vector machine model; is the regularization term coefficient of the support vector machine. Preferably, Set to 0.2.

[0096] 4. In order to enable the model to meet the requirements of incremental learning, the present invention adopts an incremental update strategy, which can update the support vector in real time as new data is added. After each new data is used, the model adjusts the weights through the following incremental update formula:

[0097]

[0098] In the formula, is the weight of the updated support vector machine model; is the weight of the support vector machine model before updating; is the learning rate of the support vector machine. Preferably, Set to 0.01.

[0099] 5. After all the above optimization steps are completed, the final support vector machine model is trained and optimized by integrating the loss functions of all steps. The final objective function is:

[0100]

[0101] In the formula, is the total loss function of the support vector machine. By minimizing this loss function, the optimal support vector machine classifier model is trained to efficiently perform data classification tasks.

[0102] (5) Data visualization and analysis module

[0103] The function of the data visualization and analysis module is to convert complex data results and model outputs into easy-to-understand graphical information to help users make decision analysis.

[0104] The functions of the data visualization and analysis module include: real-time monitoring display, visualizing real-time production data, model prediction results and process status through dashboards, etc.; data trend analysis, displaying data change trends through charts, heat maps, etc., to help users identify potential problems; prediction result display, presenting the prediction results of the machine learning model (such as equipment failure prediction, production capacity prediction, etc.) in an intuitive way, so that production personnel can respond in a timely manner.

[0105] Compared with the prior art, the present invention adopts a neural network algorithm based on gradient accumulation compensation, which can effectively avoid the gradient explosion and gradient vanishing problems encountered by traditional neural networks in large-scale process data training. Through gradient accumulation and dynamic compensation, the model can stably update parameters and accelerate the convergence process of complex process data, and has stronger adaptability and robustness.

[0106] A method for calculating sample importance adjustment and local gradient offset is proposed to weight rare and important samples to improve the model's recognition ability of these samples, especially in fault diagnosis in equipment maintenance systems. The model can pay more attention to and learn those rare but extremely important failure modes.

[0107] By adopting an incremental learning strategy, the model can be updated in real time based on new data, allowing the model to flexibly adjust parameters when faced with continuously changing production data, avoiding the limitations of a fixed training model.

[0108] A redundancy elimination mechanism based on information entropy is adopted. By calculating the correlation between features, the feature space is dynamically adjusted, redundant features are removed, and the computational efficiency and classification accuracy of the data are improved. This enables the model to retain the most informative features and avoid the performance degradation caused by redundant features.

[0109] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.

Claims

1. A large-scale data mining method for manufacturing process, characterized in that: include: Data collection, wherein the data is data acquired during the manufacturing process; Extracting features from the data, wherein a neural network algorithm based on gradient accumulation compensation is used as a feature extraction model; Classifying the data according to the features, wherein a classification model is established using a machine learning algorithm to identify the data; Storing and managing the classified data; The data are visualized.

2. A large-scale data mining method for manufacturing process as claimed in claim 1, characterized in that: The training method of the neural network algorithm based on gradient accumulation compensation includes: Initializing the parameters of the neural network, wherein the parameters of the neural network are initialized by means of a random normal distribution, and the initial value of the gradient accumulation is set to zero; In the back propagation process of each training, the gradient accumulation strategy is adopted to add and accumulate the current gradient and the historical gradient, and smoothly update it through the dynamic compensation factor; the calculation method of the gradient accumulation strategy is expressed as: In the formula, For the The gradient accumulation of iterations, For the The gradient accumulation of iterations, is a positive integer, is the gradient accumulation attenuation factor, The loss function of the neural network is parameterized The gradient of For the training samples; For the The labels of the training samples; is the loss function of the neural network; For the The local gradient offset for the iteration; are neural network parameters, which include weights and bias parameters of the neural network; Local gradient offset calculation, wherein the local gradient offset calculation performs weighted processing on important samples through sample weighting coefficients; Use the neural network's loss function to update the neural network parameter gradient; During the feedforward calculation of the neural network, the activation function is dynamically adjusted, and the activation output is combined with the current accumulated gradient information and the scale of the input sample features; Dynamically adjust the learning rate according to the gradient changes during training; Perform global updates on neural network parameters; Repeat the above steps and stop training if the continuous change of the loss function is less than the given threshold.

3. A large-scale data mining method for manufacturing process as claimed in claim 2, characterized in that: The activation function is dynamically adjusted and calculated as follows: In the formula, For the The activation value of a sample in the neural network, is a positive integer, is the Sigmoid activation function, is the number of neurons in the neural network, is the first forward propagation hyperparameter, is the second forward propagation hyperparameter, For the The cumulative gradient of the iterations, For the training samples, is the norm of the global features in the training dataset, is the L2 norm, The neural network weights, is a positive integer, The neural network A bias.

4. A large-scale data mining method for manufacturing process as claimed in claim 2, characterized in that: The adaptive learning rate adjustment strategy based on batch gradient norm statistics dynamically controls the adaptive learning rate; the calculation method of the adaptive learning rate adjustment strategy based on batch gradient norm statistics is expressed as: In the formula, For the neural network The learning rate of the iteration, is the initial learning rate of the neural network, is the number of samples input to the neural network in the current batch, is a positive integer, is the learning rate adjustment factor, is the threshold for learning rate adjustment, For the The gradient accumulation for iterations.

5. The large-scale data mining method for manufacturing process according to claim 2, characterized in that: The update method of the loss function of the neural network to the parameter gradient is expressed as: In the formula, is the parameter update operation, is the gradient update adjustment factor, is the preset gradient threshold, is a hyperparameter that controls the nonlinear scaling of the gradient.

6. A large-scale data mining method for manufacturing process as claimed in claim 2, characterized in that: The classification model structure is based on a support vector machine algorithm.

7. A large-scale data mining method for manufacturing process as claimed in claim 1 or 2, characterized in that: The data acquired in the manufacturing process includes production equipment sensor data, production process monitoring system data, equipment maintenance system data, quality inspection equipment data, and manually recorded data.

8. A large-scale data mining system for manufacturing processes, characterized in that: include: A data acquisition module, wherein the data acquisition module is used to acquire data in the manufacturing process; A data processing module, the data processing module is used to extract features from the data; A data classification module, wherein the data classification module is used to classify the data according to the features; Data storage and management module: The data storage and management module is used to store and manage the classified data; A data visualization and analysis module is used to process and output the data visualization.

9. A computer-readable storage medium storing instructions, characterized in that: The storage medium stores a computer program or instruction, and when the computer program or instruction is executed by the image processing device, the method as claimed in any one of claims 1 to 6 is implemented.

10. A computer program product, characterized in that The computer program product comprises a computer program code, and when the computer program code is executed, a processor is caused to execute any one of the methods of claims 1-6.

Images