Fault level assessment method for production line equipment in intelligent manufacturing industry based on big data

Abstract

The present invention relates to the field of artificial intelligence technology, and discloses a method for evaluating the fault level of production line equipment in the intelligent manufacturing industry based on big data, including obtaining equipment operation data; constructing and training a large model for equipment fault identification, inputting the equipment operation data into the trained large model for equipment fault identification, and outputting equipment fault cause data; constructing and training an equipment fault level evaluation model, and inputting the equipment operation data and equipment fault cause data into the trained equipment fault level evaluation model after natural language processing, and outputting equipment fault level evaluation results. The present invention significantly improves the accuracy of fault diagnosis and the accuracy of fault level determination by performing deep feature extraction and effective classification on fault data, thereby improving the operating efficiency of the equipment and the timeliness of maintenance.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method for evaluating the fault level of production line equipment in the intelligent manufacturing industry based on big data. Background Art

[0002] In the process of implementing intelligent manufacturing, real-time monitoring and fault prediction of equipment are the key to ensuring production efficiency and product quality. However, existing technologies have certain limitations in processing complex equipment operation data and accurately diagnosing equipment failures, especially in extracting useful information from massive production data and identifying and predicting potential equipment failure levels.

[0003] The Chinese invention patent with application number CN202410718037.1 proposes an optimization method and an optimization system for training samples based on bidirectional iteration of a model, which relates to the field of sample training of a model. In order to solve the defects of the existing small model, such as poor expression ability, inability to understand complex problems, slow processing speed and high cost of the large model, the present invention inputs the prompt word library into the large model to output a label set; generates a training sample set; inputs the label set and the total training sample set into the downstream small model for task training to obtain a downstream task model; combines the real data corresponding to the verification result error with the training sample set, filters the inferior data, and optimizes the training sample set; repeats the optimization until the pass rate reaches the preset threshold to complete the first data quality optimization iteration, increases the number of training samples, and repeats the optimization iteration until the verification pass rate of all labeled real data reaches the preset threshold, and obtains a verified downstream task model. The present invention is mainly used for the training sample method of the optimization model.

[0004] The Chinese invention patent with application number CN202410652558.1 proposes a large model fine-tuning method, device, equipment and medium based on supervised fine-tuning, which is applied to large language models and involves the field of large models, including: generating corresponding synthetic data based on the large language model, and constructing a training data set of the large language model based on the synthetic data; cleaning the training data set according to preset data cleaning rules, and annotating and verifying the cleaned data set according to preset data verification rules to obtain a target data set; fine-tuning the large language model using a preset model fine-tuning tool according to the target data set until a target large language model that meets the preset training requirements is obtained. By introducing synthetic data, a large amount of data can be generated in a short time in scenarios where there is little real data, effectively reducing the time for collecting real data and improving the efficiency of model fine-tuning.

[0005] The Chinese invention patent with application number CN202410587899.5 proposes a nonlinear fine-tuning method, device, equipment and medium for a large model, which belongs to the field of artificial intelligence technology. The nonlinear fine-tuning method for a large model includes: obtaining a pre-trained model and a data set; connecting the first bypass and the second bypass in parallel to the pre-trained model to obtain a large model; the first bypass includes a first matrix and a second matrix; the second bypass includes a third matrix, a nonlinear activation function and a fourth matrix; initializing the first bypass and the second bypass to obtain an initial model; according to the relationship between the performance parameters of the initial model and the first threshold and the second threshold, adjusting the parameters of each matrix in the first bypass and the second bypass, and using the data set to train the initial model to fine-tune the initial model to obtain the final model. The present invention efficiently improves the performance and processing capabilities of the model.

[0006] The existing technology has the following problems that still need to be further solved:

[0007] (1) In the existing technology, fault diagnosis mostly relies on traditional data processing and analysis methods, which lack the ability to deeply analyze the complex log data of production line equipment in the intelligent manufacturing industry, resulting in inaccurate identification of fault causes.

[0008] (2) In the existing technologies, parameter optimization methods are mostly based on simple gradient descent or its variants. These methods fail to fully utilize the principles of energy transfer and population dynamics in ecosystems, resulting in limited model training efficiency and parameter optimization effects.

[0009] (3) Existing technologies usually use simple linear models or shallow neural networks for feature extraction, which limits the ability to extract deep and complex features from a large amount of equipment operation data, affecting the accuracy of fault level assessment and the comprehensiveness of detail identification. Summary of the invention

[0010] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for evaluating the fault level of production line equipment in the intelligent manufacturing industry based on big data.

[0011] In order to achieve the above-mentioned object of the invention, the technical solution adopted by the present invention is:

[0012] A method for evaluating the fault level of production line equipment in the intelligent manufacturing industry based on big data comprises the following steps:

[0013] Obtain equipment operation data;

[0014] Build and train a large model for equipment fault identification, input equipment operation data into the trained large model for equipment fault identification, and output the equipment fault cause data;

[0015] An equipment fault level assessment model is constructed and trained. The equipment operation data and equipment fault cause data are processed by natural language and then input into the trained equipment fault level assessment model to output the equipment fault level assessment result.

[0016] Furthermore, the large model for equipment fault identification adopts the Transformer network.

[0017] Furthermore, the training method of the feedforward network in the equipment fault identification large model includes:

[0018] Initialize the initial population of feedforward network parameters and assign an initial energy value to each parameter; the initial energy of each parameter is calculated as follows:

[0019]

[0020] in, For the The initial energy of the parameters, For the Initial estimates of the parameters;

[0021] Calculate the energy of each parameter under the current network performance, the calculation method is:

[0022]

[0023] in, For the The iteration The energy of the parameters, is the attenuation factor, For the The iteration The energy of the parameters, is the loss function About parameters The partial derivative of is the sensitivity analysis function, The data set input for the current batch;

[0024] The interaction function is established according to the energy of each parameter under the current network performance, which is expressed as:

[0025]

[0026] in, For the The interaction function of the parameters, is the energy exchange coefficient, For the The initial estimates of the parameters, For the The iteration The initial energy of the parameters, For the The iteration The initial energy of the parameters, is the hyperbolic tangent function;

[0027] Energy transfer and redistribution are performed according to the interaction function, which is expressed as:

[0028]

[0029] in, For the The iteration The initial estimates of the parameters, For the The iteration The initial estimates of the parameters, is the learning rate, For the The interaction function of the parameters, is the modulation frequency, For the The number of iterations;

[0030] Repeat the above steps until the preset stop iteration condition is met, and the model training is completed.

[0031] Furthermore, the sensitivity analysis function is calculated as follows:

[0032]

[0033] in, The total number of data entered for the current batch, For the The network output for each data point.

[0034] Furthermore, the equipment fault level assessment model includes an equipment fault feature extraction model and an assessment classifier model.

[0035] Furthermore, the equipment fault feature extraction model adopts a multi-layer fully connected neural network.

[0036] Furthermore, the training method of the equipment fault feature extraction model includes:

[0037] The model parameters are initialized using the initial condition setting method based on quantum state representation, which is expressed as:

[0038]

[0039] in, are the initialized model parameters, is the standard deviation of the parameter initialization, and is an independent and uniformly distributed random variable in the interval [0, 1];

[0040] In the forward propagation process, the input feature data is encoded through the quantum state activation function, which is expressed as:

[0041]

[0042] in, is the output of the quantum state activation function, and are the mean and standard deviation in the activation function, respectively. To input feature data, is an imaginary unit, is the phase function;

[0043] The loss function based on quantum information entropy calculates the deviation between the output layer feature representation and the target feature. The calculation method is:

[0044]

[0045] in, is the loss function based on quantum information entropy, For the The target probability distribution of samples is For the The probability distribution of the model prediction for samples, is the regularization parameter, The number of samples entered for the current batch;

[0046] The quantum state function and chaos parameters are dynamically adjusted according to the model performance, which can be expressed as:

[0047]

[0048]

[0049] in, and In the The mean and standard deviation parameters for the iterations, and In the The mean and standard deviation parameters for the iterations, and Respectively The mean and standard deviation parameter adjustments calculated based on model performance feedback for each iteration, and is the adjustment rate;

[0050] Repeat the above steps until the preset stop iteration condition is met, and the model training is completed.

[0051] Furthermore, the extreme learning machine is used to evaluate the classifier model.

[0052] Furthermore, the training method for evaluating the classifier model includes:

[0053] Initialize the parameters of the extreme learning machine, expressed as:

[0054]

[0055]

[0056] in, is the connection weight of the extreme learning machine, is the bias of the extreme learning machine, is the number of input nodes of the extreme learning machine, is the number of output nodes of the extreme learning machine, For Random numbers uniformly distributed in the interval;

[0057] Through forward propagation, the output value of each category is calculated as follows:

[0058]

[0059] in, is the output vector of the extreme learning machine, is the Sigmoid activation function, is the input data vector of the extreme learning machine;

[0060] Compare the result of the forward propagation with the label and calculate the classification error for each data point as follows:

[0061]

[0062] in, is the classification error between the output of the extreme learning machine and the label, For labels;

[0063] The learning rate is adjusted dynamically according to the size and change trend of the error. The adjustment method is:

[0064]

[0065]

[0066] in, is the learning rate of the extreme learning machine, is the classification error between the output of the extreme learning machine of the previous iteration and the label, represents the difference between the error of the current iteration and the previous iteration, Parameters for adjusting the reaction speed;

[0067] Calculate the gradient of the error function with respect to the weight, expressed as:

[0068]

[0069] in, is the gradient of the extreme learning machine weight, is the derivative of the Sigmoid activation function;

[0070] The gradient projection method is used to adjust the gradient direction, which is expressed as:

[0071]

[0072] in, is the update increment of the current iteration of the extreme learning machine weight, is the gradient projection function;

[0073] Update the weights and biases of the extreme learning machine according to the adjusted gradient and learning rate. The update method is expressed as:

[0074]

[0075]

[0076] in, is the updated connection weight, is the updated bias;

[0077] Repeat the above steps until the preset stop iteration condition is met, and the model training is completed.

[0078] Furthermore, the gradient projection function is calculated as:

[0079]

[0080] in, is the L2 norm, To set a positive number.

[0081] The present invention has the following beneficial effects:

[0082] (1) The present invention uses a natural language generation model to achieve in-depth analysis of the equipment operation data of the production line in the intelligent manufacturing industry, which is used to accurately identify the cause of equipment failure. Combined with the multi-head self-attention mechanism and feedforward network, it can learn and extract key information from complex equipment operation logs. The present invention achieves the ability to effectively extract and identify the cause of equipment failure from high-dimensional data, and enhances the model's ability to process complex data structures.

[0083] (2) The present invention uses the ecological chain optimization algorithm to adjust the parameters of the feedforward network, simulates the energy transfer and dynamic balance mechanism in the natural ecosystem, optimizes the update process of the network parameters, and improves the efficiency of model training and the quality of parameter optimization. The present invention improves the convergence speed of model training and the accuracy of parameter update, reduces model training time, and reduces resource consumption.

[0084] (3) The fault level assessment model of the present invention uses the Word2Vec algorithm to convert the fault log text data into a numerical vector, and combines it with a fully connected neural network for in-depth feature extraction, which effectively improves the accuracy of fault diagnosis.

[0085] (4) The present invention uses a neural network algorithm based on quantum state feature constraints to deeply encode features. Through quantum state activation functions and chaotic system characteristics, the network's memory capacity and adaptability are enhanced, and the detail recognition capability of feature extraction is improved. The present invention significantly improves the accuracy of fault diagnosis and the accuracy of fault level determination by performing deep feature extraction and effective classification of fault data, thereby improving the operating efficiency of the equipment and the timeliness of maintenance.

[0086] (5) The present invention trains the classifier model based on the extreme learning machine algorithm of dynamic gradient projection, dynamically adjusts the learning rate and weight projection direction according to the gradient of the output error, so as to quickly and accurately classify data of different fault levels, thereby preventing training instability caused by gradient explosion or disappearance. BRIEF DESCRIPTION OF THE DRAWINGS

[0087] Figure 1 Flowchart of the method

[0088] Figure 2 Flowchart of the Weifang model training method DETAILED DESCRIPTION

[0089] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0090] like Figure 1 As shown, an embodiment of the present invention provides a method for evaluating the fault level of production line equipment in the intelligent manufacturing industry based on big data, comprising the following steps S1 to S3:

[0091] S1. Obtain equipment operation data;

[0092] In an optional embodiment of the present invention, the data collection of step S1 is derived from the operation log of the production line equipment in the intelligent manufacturing industry. The operation log describes the working condition and fault condition of the equipment. In one embodiment, one of the data is: "2024-01-01, equipment ID 1234, equipment type is a conveyor belt machine, which makes an abnormal sound and generates severe heat during high-speed rotation."

[0093] S2. Build and train a large model for equipment fault identification, input equipment operation data into the trained large model for equipment fault identification, and output the equipment fault cause data;

[0094] In an optional embodiment of the present invention, step S2 inputs the collected equipment operation log data into a large model to obtain the cause of the equipment failure. The large model is a natural language generation large model in the field of artificial intelligence, and the natural language generation large model has basic question-answering capabilities and multi-round dialogue capabilities.

[0095] In this embodiment, based on the natural language generation big model, the training data is used to train the intelligent manufacturing industry big model. The data for training the industry big model is realized by an artificially constructed training data set. The training data set is in the form of natural language question-answer pairs, one of which is:

[0096] Question: "2024-01-01, device ID 1234, device type is a conveyor belt machine, it made an abnormal noise and heated up severely during high-speed rotation."

[0097] Answer: "This phenomenon is caused by wear or damage to the bearings of the conveyor belt. It is recommended to stop the machine immediately to check the bearing condition and replace it if necessary."

[0098] This embodiment uses an artificially constructed training data set to train a large natural language generation model. Specifically, the large natural language generation model is a Transformer network. The Transformer network consists of an encoder and a decoder. The encoder maps input features into a series of context-aware vector representations, and the decoder gradually decodes these vector representations into target features. Its architecture is as follows: Figure 2 shown.

[0099] The encoder part is composed of The encoder is composed of 2 identical encoding layers, each of which consists of two sub-layers: a multi-head attention mechanism sub-layer and a feed-forward network sub-layer. The encoder adds residual connections and normalization operations after each sub-layer.

[0100] The multi-head attention mechanism sublayer is implemented by the multi-head self-attention mechanism, which is the core module of the encoder and is responsible for learning the importance of each position from the input features and combining the information of all positions to obtain a global contextual representation, thereby improving the performance and generalization ability of the model. Specifically, the multi-head self-attention mechanism is implemented as follows: the input features pass through three different linear transformation layers to generate "query", "key" and "value" vectors respectively, and the linear transformation is achieved by multiplying the weight matrix; further, the generated query, key and value vectors are divided into multiple smaller parts, namely "heads", so that the model processes multiple different representation subspaces in parallel, thereby capturing the diversity information in the input data; further, for each head, the dot product of the query and key vectors is calculated respectively to obtain an attention score, which reflects the relative importance of different positions in the sequence, and in order to make the numerical value of the attention score stable, the score is divided by a scaling factor, and the scaling factor can be the square root of the key vector dimension; further, the attention score is processed by the Softmax function to obtain the attention weight of each position, and the Softmax function ensures that all weights add up to 1, so that it can be processed as a probability distribution, expressed as:

[0101]

[0102] in, For the The attention weight of each head, For the The query vector of the head, For the The key vector of the head, For the The transpose of the key vector of the head, is the dimension of the input features, is the number of attention heads, For the A vector of values ​​for the heads.

[0103] This embodiment uses the obtained attention weights to perform weighted summation on the value vector to obtain an output vector; further, the output vectors of all heads are concatenated together, and then the output of the final multi-head self-attention layer is generated through a linear transformation layer.

[0104] The feedforward network sublayer consists of two fully connected layers and an activation function, which is used to perform nonlinear transformation on the vector at each position. In this sublayer, the input vector first passes through the first fully connected layer for linear transformation, then passes through a ReLU activation function for nonlinear transformation, and finally passes through the second fully connected layer for linear transformation to improve the expressiveness and generalization ability of the model, so that the model can better handle complex input data.

[0105] The decoder is The decoder layer is composed of the same decoding layers. Unlike the encoding layer, each decoding layer has an additional multi-head attention mechanism sublayer, that is, the decoding layer contains two multi-head attention mechanism sublayers and one feedforward network mechanism sublayer. In addition, the second multi-head attention mechanism sublayer of the decoding layer is implemented by a multi-head cross attention mechanism, which is a mechanism for information exchange between the encoder and the decoder. It takes the output features of the encoder and the input features of the decoder as input, and uses the attention mechanism to perform cross-attention calculations on the input, thereby obtaining a set of weighted decoder output features.

[0106] It should be noted that in the encoder and decoder of the Transformer network, the output of each sub-layer is the result of layer normalization and residual connection, expressed as:

[0107]

[0108] in, is the layer normalization function, represents the output of the residual connected sublayer, Represents the input of the sublayer.

[0109] It should also be noted that the feedforward network of the Transformer network is a neural network structure composed of neurons, which plays a role in enhancing the nonlinear processing ability of the model. The neural network structure includes weights and bias parameters. In order to avoid the gradient disappearance and gradient descent phenomenon caused by the traditional neural network using the gradient descent method to update the parameters, and to avoid the phenomenon that the parameters fall into the local optimal solution after the training is completed, the present invention adopts the ecological chain optimization algorithm to update the parameters of the feedforward network during the training process of the feedforward network, which is inspired by the ecosystem in nature, especially the energy flow and population dynamic balance mechanism in the food chain. In the ecosystem in nature, different species influence each other through the food chain and jointly maintain the ecological balance. The interaction and energy transfer between species provide an efficient and adaptive adjustment mechanism for the ecosystem. The ecological chain optimization algorithm simulates the energy transfer and population control mechanism in the ecosystem, and is applied to the parameter optimization of the feedforward network. Each feedforward network parameter is regarded as a species in the ecosystem, and its energy represents the effectiveness of the parameter, that is, the contribution to the final output. Unlike the traditional gradient descent algorithm, the ecological chain optimization algorithm allows the parameters to self-adjust during the optimization process, and can balance the global search with the local refined search, thereby avoiding falling into the local optimal solution.

[0110] The training process of the feedforward network based on the ecological chain optimization algorithm in this embodiment is as follows:

[0111] (1) Initialize and set the initial population of feedforward network parameters, and assign an initial energy value to each parameter. In one embodiment, the network has parameters, the calculation method of the initial energy of each parameter is expressed as:

[0112]

[0113] in, It is The initial estimates of the parameters, For the The initial energy of the parameters. Preferably, It follows a normal distribution with a mean of 0.01 and a variance of 0.01.

[0114] (2) Calculate the energy contribution of each parameter under the current network performance. A high energy contribution indicates a good parameter configuration effect. Specifically, In the iterations, the parameters The energy update method is expressed as:

[0115]

[0116] in, is the loss function About parameters The partial derivative of For the The iteration The energy of the parameters, For the The iteration The energy of the parameters, is the sensitivity analysis function, The data set input for the current batch, is the attenuation factor. Preferably, Set to 0.9.

[0117] In one embodiment, the sensitivity analysis function is used to quantify the parameter The response sensitivity for different data inputs is calculated as follows:

[0118]

[0119] in, It is for The network output for each data point is Is the total number of data input in the current batch.

[0120] (3) Simulate the relationship between predators and herbivores to adjust parameters. According to the food chain level, simulate the predation process of higher-level parameters (predators) on lower-level parameters (herbivores) and adjust their energy (i.e. parameter value). Specifically, set the parameter and At different levels of the food chain, the interaction function is defined as:

[0121]

[0122] in, is the interaction function, is the energy exchange coefficient, It is The initial estimates of the parameters, For the The iteration The initial energy of the parameters, For the The iteration The initial energy of the parameters, is a hyperbolic tangent function. Preferably, Set to 0.01.

[0123] (4) Perform energy transfer and redistribution, that is, transfer part of the energy lost by the prey parameter to the predator to achieve energy redistribution and ensure the improvement of the overall performance of the network. The calculation method is expressed as:

[0124]

[0125] in, is the learning rate, It is The iteration The initial estimates of the parameters, It is The iteration The initial estimates of the parameters, is the modulation frequency, is the number of iterations of the current iteration. Preferably, Set to 0.01, Set to 1.

[0126] (5) Repeat the above steps until the preset stop iteration condition is met, indicating that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0127] S3. Build and train an equipment fault level assessment model, perform natural language processing on the equipment operation data and equipment fault cause data, and input them into the trained equipment fault level assessment model to output the equipment fault level assessment result.

[0128] In an optional embodiment of the present invention, step S3 inputs the collected equipment operation log data into the large model. After obtaining the cause of the equipment failure, the equipment operation log and the cause of the failure are used together as training data for the equipment failure level assessment model to train the equipment failure level assessment model.

[0129] Specifically, the Word2Vec algorithm in natural language processing technology is used to convert each word in the equipment operation log and the cause of the fault into a numerical vector, and then the vector representation of the entire data is formed by vector averaging, which is recorded as .

[0130] Furthermore, the vector representation of the data The labeling method of the present invention is manual labeling, that is, the vector representation of the data is manually Assign labels. In one embodiment, the categories of labels include: "early stage of equipment failure", "mid-stage of equipment failure", and "late stage of equipment failure".

[0131] The equipment fault level assessment model constructed in this embodiment includes an equipment fault feature extraction model and an assessment classifier model.

[0132] When training the equipment fault feature extraction model, the training data of the equipment fault level assessment model is input into the feature extraction model to train the feature extraction model. This embodiment uses a 6-layer fully connected neural network for feature extraction. In the prior art, some solutions use neural networks for feature extraction. In some neural network structures, gradient vanishing, gradient explosion, or falling into local optimal solutions may be encountered, affecting the stability of training and the performance of the model. This embodiment uses a neural network algorithm based on quantum state feature constraints as a feature extraction model. On the basis of traditional neural networks, the dynamic system characteristics of chaos theory are used to increase the memory capacity and adaptability of the network through nonlinear dynamic nodes, thereby extracting features more accurately.

[0133] The training process of the neural network algorithm based on quantum state feature constraints in this embodiment is as follows:

[0134] (1) Initializing the parameters of the neural network model. The present invention adopts an initial condition setting method based on quantum state representation. In one embodiment, the initialization method is expressed as:

[0135]

[0136] in, Represents the parameters of the initialized neural network model, including the weights and biases of the neural network model; is the standard deviation of the parameter initialization; and is an independent and uniformly distributed random variable in the interval [0, 1]; The cosine function is used to adopt periodic changes to simulate the sensitive dependence of the system state in chaos theory. In one embodiment, Set to 0.3.

[0137] (2) During the forward propagation process, the input feature data is first transformed nonlinearly through the quantum state activation function, and then processed layer by layer to adjust the activation state of the nodes in each layer, thereby performing deep encoding of the features. The calculation method of the quantum state activation function of the neurons in each layer of the neural network is expressed as:

[0138]

[0139] in, is the input feature of the neural network; is the output of the quantum state activation function; and They are the mean and standard deviation in the activation function, which are used to adjust the center and spread of data distribution; is an imaginary unit; is the phase function.

[0140] In one embodiment, the phase function is calculated based on the quantum behavior characteristics of the input data, so that it includes a nonlinear response to the data characteristics and the dynamic characteristics of the chaotic system. The calculation method is expressed as:

[0141]

[0142] in, is the main frequency parameter of the quantum phase; Adjust the sensitivity of nonlinear response; is the hyperbolic tangent function; is the parameter related to the chaotic characteristic amplitude; is the frequency parameter of the chaotic system; is a sine function. Preferably, Set to 2, Set to 3.

[0143] (3) In order to simultaneously consider the fidelity and collapse risk of the quantum state, the deviation between the output layer feature representation and the target feature is calculated by a loss function based on quantum information entropy. The calculation method is expressed as:

[0144]

[0145] in, It is a loss function based on quantum information entropy; It is The target probability distribution of samples; It is The probability distribution of the model predictions for samples; is the regularization parameter; is the number of samples input in the current batch.

[0146] The loss function based on quantum information entropy is used to update the parameters, and the update method is expressed as:

[0147]

[0148] in, and They represent the neural network in The iteration and The parameter value for the iteration; is the learning rate of the neural network; is the quantum noise vector; is the loss function About parameters Preferably, Set to 0.01.

[0149] In one embodiment, the quantum noise vector is dynamically adjusted according to the model performance, and the calculation method is expressed as:

[0150]

[0151] in, is the noise intensity factor; is the sensitivity parameter used to adjust the noise intensity; It is The loss value of samples; is the expected loss value. Preferably, Set to 0.3, Set to 0.05, Set to 0.1.

[0152] gradient The calculation method is expressed as:

[0153]

[0154] (4) Dynamically adjust the quantum state function and chaos parameters according to the performance of the model. The adjustment method is expressed as:

[0155]

[0156]

[0157] in, and In the The mean and standard deviation parameters for the iterations; and In the The mean and standard deviation parameters for the iterations; and It is The adjustment amount calculated based on the model performance feedback for the iteration; and is the adjustment rate. Preferably, and Set to 0.01 and 0.03 respectively.

[0158] Adjustment amount and The calculation method is expressed as:

[0159]

[0160] in, It is to adjust the reaction rate; and The data sets are The mean and variance of the iterations. Preferably, Set to 0.5.

[0161] (5) Repeat the above steps until the preset stop iteration condition is met, indicating that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0162] When training the evaluation classifier model, the feature-extracted data is input into the evaluation classifier model for training the evaluation classifier model. This embodiment adopts an extreme learning machine algorithm based on dynamic gradient projection as a classification algorithm. On the basis of the traditional extreme learning machine, a dynamic weight rebalancing mechanism and a gradient projection strategy are adopted to adjust the connection weights in real time, and the learning rate and weight projection direction are dynamically adjusted according to the gradient of the output error, thereby improving the classification accuracy and convergence speed.

[0163] The training process of the extreme learning machine algorithm based on dynamic gradient projection in this embodiment is as follows:

[0164] (1) Initialize the parameters and weights of the extreme learning machine and bias The initialization method is expressed as:

[0165]

[0166]

[0167] in, represents the connection weight of the extreme learning machine; represents the bias of the extreme learning machine; is the number of input nodes of the extreme learning machine; is the number of output nodes of the extreme learning machine; Indicated in A uniformly distributed random number in the interval.

[0168] (2) The data after input feature extraction is forward propagated to calculate the output value of each category. Specifically, for a given input vector, the calculation method of the output of forward transmission is expressed as:

[0169]

[0170] in, is the Sigmoid activation function; is the input data vector of the extreme learning machine; is the output result vector of the extreme learning machine.

[0171] (3) Compare the result of the forward propagation with the actual label and calculate the classification error of each data point. The calculation method is expressed as:

[0172]

[0173] in, is the error between the output of the extreme learning machine and the actual label.

[0174] The learning rate is adjusted dynamically according to the size and change trend of the error. The adjustment method is expressed as:

[0175]

[0176]

[0177] in, is the learning rate of the extreme learning machine, is the error between the output of the extreme learning machine of the previous iteration and the actual label, represents the difference between the error of the current iteration and the previous iteration, is a parameter for adjusting the reaction speed. Preferably, Set to 0.3.

[0178] (4) Calculate the gradient of the error function with respect to the weight, and then use the gradient projection technique to adjust the gradient direction to ensure that the weight update minimizes the error. The relationship between the gradient of the weight, the error, and the input is:

[0179]

[0180] in, is the derivative of the Sigmoid activation function, is the gradient of the extreme learning machine weight.

[0181] The gradient projection technique is used to optimize the weight update direction. The calculation method is expressed as:

[0182]

[0183] in, is the update increment of the current iteration of the extreme learning machine weight, is the gradient projection function.

[0184] To ensure that the weight update does not deviate from the reasonable range due to gradient explosion, the calculation method of the gradient projection function is expressed as:

[0185]

[0186] in, is the L2 norm; is a small positive number to prevent the denominator from being zero. Preferably, Set to 0.001.

[0187] (5) Update the weights and biases of the extreme learning machine according to the adjusted gradient and learning rate. The update method is expressed as:

[0188]

[0189]

[0190] in, is the updated connection weight, is the updated bias, Represents a parameter update operation.

[0191] (6) Repeat the above steps until the preset stop iteration condition is met, indicating that the model training is completed. In one embodiment, the preset stop iteration condition is reaching a preset maximum number of iterations. Preferably, the preset maximum number of iterations is set to 1000 times.

[0192] This embodiment uses a trained model to effectively process and classify new equipment data samples to identify possible equipment failures and operational anomalies. In one embodiment, the collected equipment operation log data is input into the large model, and after obtaining the equipment failure cause, the equipment operation log and the failure cause are vectorized and processed, and then input into the trained feature extraction model and classifier model to obtain the equipment failure level of the production line in the intelligent manufacturing industry. In this embodiment, the obtained equipment failure level may be: "Early Equipment Failure", "Mid-term Equipment Failure", "Late Equipment Failure", and based on different failure levels, corresponding equipment predictive maintenance or repair measures are taken.

[0193] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0194] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0195] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0196] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present invention.

[0197] 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 method for evaluating the fault level of production line equipment in the intelligent manufacturing industry based on big data. It is characterized in that The following steps are involved: Obtain equipment operation data; the data comes from the operation log of the production line equipment in the intelligent manufacturing industry, and the operation log describes the working condition and fault condition of the equipment; A large model for equipment fault identification is constructed and trained, equipment operation data is input into the trained large model for equipment fault identification, and equipment fault cause data is output; wherein the large model for equipment fault identification adopts a Transformer network; and the training method of the feedforward network in the large model for equipment fault identification includes: Initialize the initial population of feedforward network parameters and assign an initial energy value to each parameter; the initial energy of each parameter is calculated as follows: in, For the The initial energy of the parameters, For the Initial estimates of the parameters; Calculate the energy of each parameter under the current network performance, the calculation method is: in, For the The iteration The energy of the parameters, is the attenuation factor, For the The iteration The energy of the parameters, is the loss function About parameters The partial derivative of is the sensitivity analysis function, The data set input for the current batch; The interaction function is established according to the energy of each parameter under the current network performance, which is expressed as: in, For the The interaction function of the parameters, is the energy exchange coefficient, For the The initial estimates of the parameters, For the The iteration The initial energy of the parameters, For the The iteration The initial energy of the parameters, is the hyperbolic tangent function; Energy transfer and redistribution are performed according to the interaction function, which is expressed as: in, For the The iteration The initial estimates of the parameters, For the The iteration The initial estimates of the parameters, is the learning rate, For the The interaction function of the parameters, is the modulation frequency, For the The number of iterations; Repeat the above steps until the preset stop iteration condition is met, and the model training is completed; The calculation method of the sensitivity analysis function is as follows: in, The total number of data entered for the current batch, For the The network output for each data point; An equipment fault level assessment model is constructed and trained. The equipment operation data and equipment fault cause data are processed by natural language and then input into the trained equipment fault level assessment model to output the equipment fault level assessment result.

2. According to the big data-based intelligent manufacturing industry production line equipment fault level assessment method according to claim 1, It is characterized in that The equipment fault level assessment model includes an equipment fault feature extraction model and an assessment classifier model.

3. According to the big data-based intelligent manufacturing industry production line equipment fault level assessment method of claim 2, It is characterized in that The equipment fault feature extraction model adopts a multi-layer fully connected neural network.

4. According to the big data-based intelligent manufacturing industry production line equipment fault level assessment method of claim 3, It is characterized in that The training method of the equipment fault feature extraction model includes: The model parameters are initialized using the initial condition setting method based on quantum state representation, which is expressed as: in, are the initialized model parameters, is the standard deviation of the parameter initialization, and is an independent and uniformly distributed random variable in the interval [0, 1]; In the forward propagation process, the input feature data is encoded through the quantum state activation function, which is expressed as: in, is the output of the quantum state activation function, and are the mean and standard deviation in the activation function, respectively. To input feature data, is an imaginary unit, is the phase function; The loss function based on quantum information entropy calculates the deviation between the output layer feature representation and the target feature. The calculation method is: in, is the loss function based on quantum information entropy, For the The target probability distribution of samples is For the The probability distribution of the model prediction for samples, is the regularization parameter, The number of samples entered for the current batch; The quantum state function and chaos parameters are dynamically adjusted according to the model performance, which can be expressed as: , in, and In the The mean and standard deviation parameters for the iterations, and In the The mean and standard deviation parameters for the iterations, and Respectively The mean and standard deviation parameter adjustments calculated based on model performance feedback for each iteration, and is the adjustment rate; Repeat the above steps until the preset stop iteration condition is met, and the model training is completed.

5. According to the big data-based intelligent manufacturing industry production line equipment fault level assessment method according to claim 4, It is characterized in that The extreme learning machine is used to evaluate the classifier model.

6. According to the big data-based intelligent manufacturing industry production line equipment fault level assessment method of claim 5, It is characterized in that The training methods for evaluating the classifier model include: Initialize the parameters of the extreme learning machine, expressed as: , in, is the connection weight of the extreme learning machine, is the bias of the extreme learning machine, is the number of input nodes of the extreme learning machine, is the number of output nodes of the extreme learning machine, For Random numbers uniformly distributed in the interval; Through forward propagation, the output value of each category is calculated as follows: in, is the output result vector of the extreme learning machine, is the Sigmoid activation function, is the input data vector of the extreme learning machine; Compare the result of the forward propagation with the label and calculate the classification error for each data point as follows: in, is the classification error between the output of the extreme learning machine and the label, For labels; The learning rate is adjusted dynamically according to the size and change trend of the error. The adjustment method is: , in, is the learning rate of the extreme learning machine, is the classification error between the output of the extreme learning machine of the previous iteration and the label, represents the difference between the error of the current iteration and the previous iteration, Parameters for adjusting the reaction speed; Calculate the gradient of the error function with respect to the weight, expressed as: in, is the gradient of the extreme learning machine weight, is the derivative of the Sigmoid activation function; The gradient projection method is used to adjust the gradient direction, which is expressed as: in, is the update increment of the current iteration of the extreme learning machine weight, is the gradient projection function; Update the weights and biases of the extreme learning machine according to the adjusted gradient and learning rate. The update method is expressed as: , in, is the updated connection weight, is the updated bias; Repeat the above steps until the preset stop iteration condition is met, and the model training is completed.

7. According to the big data-based intelligent manufacturing industry production line equipment fault level assessment method of claim 6, It is characterized in that The gradient projection function is calculated as: in, is the L2 norm, To set a positive number.

Images