Fault early warning method and system for cloth production equipment

Abstract

The invention discloses a cloth production equipment fault early warning method and system, relates to the technical field of fault early warning, and aims to solve the technical problems that in the prior art, equipment fault alarm is not timely and inaccurate, and fault judgment is difficult to assist. Determining parameters according to a predetermined cloth production device and parameters required for establishing a fault alarm model; s2, fault classification and sample collection; s2, performing fault classification and corresponding sample data acquisition on the cloth production device based on the parameters determined in the step S1; s3, modeling a fault alarm model; s2, modeling a fault alarm model based on the sample data collected in the step S2; s4, judging the accuracy rate of the model; and S3, based on the fault alarm model established in the step S3, determining the accuracy of the fault alarm model and judging whether the accuracy of the model meets the requirement, and the method has the advantage of quickly and accurately judging whether the equipment has the fault and judging the fault type.

Description

Technical Field

[0001] The present invention relates to the technical field of fault warning, and more specifically, to a method and system for warning fault of fabric production equipment. Background Art

[0002] In the fabric production industry, stable operation of equipment is crucial to production efficiency and product quality. With the continuous improvement of industrial automation, fabric production equipment is becoming increasingly complex, and the losses caused by equipment failure are becoming more and more serious.

[0003] At present, the handling of fabric production equipment failures mostly relies on post-fault maintenance, that is, after the equipment fails, the staff will conduct troubleshooting and repairs based on their experience. This method has many disadvantages. On the one hand, due to the lack of real-time monitoring and early warning of the equipment's operating status, equipment failures are often difficult to detect in advance. Once a failure occurs, it may lead to production interruptions, resulting in waste of raw materials, delays in production progress, and increased production costs; on the other hand, post-fault maintenance requires a long time to troubleshoot, during which the equipment is down for a long time, which seriously affects production efficiency. In addition, during the troubleshooting process, due to the lack of effective auxiliary means, it is difficult for engineers to quickly determine the type, location and cause of the fault, which further prolongs the equipment downtime and reduces the economic benefits of the enterprise. In view of this, we propose a method and system for early warning of fabric production equipment failures. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for warning faults of fabric production equipment, so as to solve the technical problems in the prior art that equipment fault alarms are not timely and accurate, and are difficult to assist in fault judgment.

[0005] In order to solve the above technical problems, the present invention provides the following technical solutions: a method for early warning of a fault of a fabric production equipment, comprising the following steps: S1. Parameter determination; Determine parameters based on a predetermined fabric production device and parameters required to establish a fault alarm model; S2, fault classification and sample collection; Based on the parameters determined in step S1, fault classification of the fabric production device and corresponding sample data collection; S3, fault alarm model building; Based on the sample data collected in step S2, a fault alarm model is built; S4, model accuracy judgment; Based on the fault alarm model established in step S3, the accuracy of the fault alarm model is determined and whether the model accuracy meets the requirements is judged. If it does not meet the requirements, step S2 is executed; if it meets the requirements, step S5 is executed; S5, device data collection; Based on the fault alarm model established in step S3, data of the fabric production device is collected; Determine whether the data collected from the current fabric production device is sufficient, if not, collect new sample data, otherwise, proceed to step S6; The collected data of the fabric production device is sent to the fault alarm model for identification. If the accuracy of the identification result is greater than the set accuracy threshold , then proceed to step S7, otherwise, further collect sample data according to step S2.

[0006] S6, fault identification; Based on the fault alarm model established in step S3, fault identification is performed on the collected data of the fabric production device; S7. Fault analysis and judgment; Based on the recognition result of step S6, whether there is a fault in the fabric production device is analyzed and judged, if there is no fault, step S5 is executed, if there is a fault, step S8 is executed; S8, alarm is issued; Based on the judgment result of step S7, an alarm instruction is sent to the alarm device via the signal calling device.

[0007] Preferably, determining the parameters of the fabric production device in S1 includes the following process: The fabric production device is abstracted into a physical diagram according to the device parameters; Establish the relationship between the physical diagram and the circuit diagram, and convert the physical diagram of the fabric production device into a circuit diagram; Establish the equivalent transformation between the physical diagram and the circuit diagram, and the rules for abstraction according to the circuit diagram are: Abstracting the fabric production device as a resistor in a circuit diagram ,capacitance and inductance The analog components, including resistors ,capacitance and inductance The parameter calculation formulas are: , is the resistivity, is the conductor length, is the cross-sectional area of ​​the conductor; , is the dielectric constant, is the plate area, is the plate spacing; , is the magnetic permeability, is the number of coil turns, is the cross-sectional area of ​​the coil, is the coil length; The connection mode between the circuit and the inductor or resistor on the fabric production device is mapped onto the circuit diagram, the connection mode between each component and the power supply is mapped onto the circuit diagram, and the logical relationship between the circuit components on the fabric production device is mapped onto the circuit diagram.

[0008] Preferably, in said S2, the fault classification of the fabric production device is divided into three types: mechanical part fault, circuit part fault and software part fault; Sample data collection for fabric production devices includes: Monitor the circuit components on the fabric production device. Suppose the monitoring time series is , at every moment The collected circuit component parameter vector is ; In the fabric production task performed by the fabric production device, whether there is a fault is used as a sample label , Indicates a fault. Indicates that there is no fault, and the parameter vector and sample label are used as sample data ; The cloth production device failures are divided into training sets according to whether the sample data is sufficient. and test set , assuming the total number of samples is ,like , then randomly select samples as training set , the remaining samples are used as the test set , where the number of training set samples is and the number of test set samples satisfy ,and ,in, is the preset sufficient sample number threshold, ; Divide the fabric production task execution time into several time periods , in each time period Extract a batch of sample data , select part of the sample data as training sample data, and the rest of the sample data as test sample data. According to the fault conditions and corresponding sample labels, a fault condition database is established for the subsequent training of the fault alarm model. Set the time period The total number of sample data drawn from , select the ratio , then the number of training samples , the number of test samples ,in, .

[0009] Preferably, in said S3, modeling a fault alarm model of a fabric production device includes the following process: Establish a fault alarm model based on convolutional neural network, which includes convolutional network layer, pooling layer and fully connected layer; In the convolutional network layer, convolution kernels are used to extract features from the input sample data. Each convolutional network layer includes several feature maps and outputs the extracted features. Suppose the input sample data is , the convolution kernel is , the convolution step size is , then the convolutional layer outputs features The calculation formula is: , in, is the convolution kernel size, is the number of input data channels, is the bias term, , , are the row, column, and channel indices of the output feature map, respectively. Represents the position index of the element in the convolution kernel and the number of feature maps output by each convolutional network layer The number of feature maps in the previous layer and the number of convolution kernels The relationship is ; In the pooling layer, the pooling method is used to compress the features extracted by the convolutional network layer, reduce data redundancy, extract features, reduce prediction errors, and improve prediction accuracy. The pooling window size is set to , then the pooling layer output The calculation formula is: , in, The range is , , , They are the row, column, and channel indices of the output feature map, respectively; In the fully connected layer, the ReLU activation function is used to combine the data output by the convolutional network layer and the pooling layer to output the prediction result. Suppose the input of the fully connected layer is , the weight matrix is , the bias vector is , then the fully connected layer outputs The calculation formula is ,in, , Represents the original data.

[0010] Preferably, in S4, the accuracy determination of the model includes the following process: The training sample data is distributed to the convolution kernel of the convolutional neural network according to the preset ratio. Suppose the total number of training sample data is , assigned to The number of samples of the convolution kernel is ,but ,in, is the number of convolution kernels; Set the convolutional neural network structure and set the hyperparameters, including the number of iterations , error function , the error function uses the cross entropy loss function ,in, is the true label of the sample, Predict output for the model; The training sample data is input into the convolutional neural network for training to obtain the convolutional neural network training results. During the training process, the model parameters are updated through the back propagation algorithm. The weight update amount is set and bias update The calculation formulas are , ,in, is the learning rate; Input the test sample data into the training results to obtain the predicted data, and compare the predicted data with the labels corresponding to the sample data; Setting the accuracy threshold The accuracy threshold is 80%, and the accuracy of the fault alarm model is determined based on the comparison results. , let the number of samples predicted correctly be , then the accuracy ,like , execute S5, if , execute S2.

[0011] Preferably, in S6, fault identification is performed based on the fault alarm model, including the following process: The collected data of the fabric production device is transmitted as input to the fault alarm model, feature extraction is performed based on the convolutional network layer in the fault alarm model, feature compression is performed based on the pooling layer in the fault alarm model, and the prediction result is output after fusion based on the fully connected layer in the fault alarm model.

[0012] Preferably, in S7, analyzing and judging whether the device has a fault includes the following process: Analyze the characteristic information and data relationships of the sample data, including determining the number of data collected by the device , determine the frequency of sample data collection , fault identification is performed through the fault alarm model, and the fault probability output by the model is assumed to be ,but Calculated by the fault alarm model in step S3; According to the data distribution and model recognition effect, the probability value of the model recognition result is determined, and this value is set as the recognition threshold of the model. , let the mean of the data distribution be , the standard deviation is ,but ,in, is a coefficient determined empirically; Perform input data preprocessing for the fault alarm model, including data normalization and scaling. Suppose the original data is The normalized data is , then the normalization formula is , the scaling formula is ,in, is the scaling factor, is the offset; Input the preprocessed data into the fault alarm model, including setting the input layer, hidden layer and output layer of the model and determining the input data and output data. The input data is the preprocessed sample data, and the output data is the fault prediction probability. The recognition results of the data calculated by the fault alarm model include the recognition probability of the set output , comparing the recognition probability with the recognition threshold, if the recognition probability is greater than the threshold, it is considered that there is a fault, otherwise there is no fault; Determine whether there is a fault. If there is no fault, go to step S5; if there is a fault, go to step S8.

[0013] A cloth production equipment fault early warning system, comprising: A device parameter identification module, used for establishing parameters required to set a fault alarm model according to a predetermined fabric production device; A sample collection module, used to collect sample data according to the parameters determined by the device parameter identification module; A model building module is used to build a fault alarm model based on the collected sample data; The accuracy judgment module is used to judge whether the accuracy of the model meets the requirements according to the established fault alarm model; A real-time data acquisition module is used to collect data of the material distribution device in real time according to the established fault alarm model; A fault identification module, used for performing fault identification based on the collected sample data; The alarm module is used to generate an alarm based on the identification result of the fault identification module.

[0014] Compared with the prior art, the present invention has the following beneficial effects: 1. The present invention can predict possible equipment failures in advance by real-time monitoring and analysis of the operating status and parameters of the fabric production equipment. Before the failure occurs, an alarm is issued in time according to the model calculation results to remind relevant personnel to take measures, thereby effectively improving the safety performance of the equipment, avoiding production interruptions, casualties and property losses caused by sudden equipment failures, ensuring the continuity and stability of production, and solving the problems of untimely and inaccurate equipment failure alarms and difficulty in assisting fault judgment in the prior art.

[0015] 2. The present invention can more accurately determine whether there is a fault in the equipment and the type of fault through feature analysis of sample data, scientific determination of fault identification thresholds, data preprocessing, and the use of a fault alarm model based on a convolutional neural network. It can assist engineers in quickly locating the fault site and cause. This allows timely maintenance and adjustments to be made during the production process once a potential fault or failure is discovered, thus minimizing equipment downtime, quickly restoring equipment to normal operation, improving production efficiency, and reducing the negative impact of equipment failure on production.

[0016] 3. The present invention sets a strict process for judging the model accuracy, including allocating the training sample data to the convolution kernel in proportion, setting the convolutional neural network structure and hyperparameters, calculating the error through the cross-entropy loss function, updating the model parameters using the back propagation algorithm, and setting the accuracy threshold. Only when the model accuracy meets the requirements will subsequent equipment data collection and fault identification be carried out. This method effectively avoids the situation where false alarms are issued due to inaccurate model recognition rate, and ensures the accuracy of the equipment fault alarm identification results. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a schematic diagram of the process of the present invention; Figure 2 Schematic diagram of the system framework of the present invention. DETAILED DESCRIPTION

[0018] In order to facilitate those skilled in the art to understand the technical solution of the present invention, the technical solution of the present invention is further described in conjunction with the accompanying drawings of the specification.

[0019] Embodiment 1, as Figure 1 As shown, the present invention provides a method for early warning of a fault of a cloth production device, comprising the following steps: S1. Parameter determination; Determine parameters based on a predetermined fabric production device and parameters required to establish a fault alarm model; S2, fault classification and sample collection; Based on the parameters determined in step S1, fault classification of the fabric production device and corresponding sample data collection; S3, fault alarm model building; Based on the sample data collected in step S2, a fault alarm model is built; S4, model accuracy judgment; Based on the fault alarm model established in step S3, the accuracy of the fault alarm model is determined and whether the model accuracy meets the requirements is judged. If it does not meet the requirements, step S2 is executed; if it meets the requirements, step S5 is executed; S5, device data collection; Based on the fault alarm model established in step S3, data of the fabric production device is collected; Determine whether the data collected from the current fabric production device is sufficient, if not, collect new sample data, otherwise, proceed to step S6; The collected data of the fabric production device is sent to the fault alarm model for identification. If the accuracy of the identification result is greater than the set accuracy threshold , then proceed to step S7, otherwise, further collect sample data according to step S2.

[0020] S6, fault identification; Based on the fault alarm model established in step S3, fault identification is performed on the collected data of the fabric production device; S7. Fault analysis and judgment; Based on the recognition result of step S6, whether there is a fault in the fabric production device is analyzed and judged, if there is no fault, step S5 is executed, if there is a fault, step S8 is executed; S8, alarm is issued; Based on the judgment result of step S7, an alarm instruction is sent to the alarm device via the signal calling device.

[0021] In an embodiment of the present invention, determining the parameters of the cloth production device in S1 includes the following process: The fabric production device is abstracted into a physical diagram according to the device parameters; Establish the relationship between the physical diagram and the circuit diagram, and convert the physical diagram of the fabric production device into a circuit diagram; Establish the equivalent transformation between the physical diagram and the circuit diagram, and the rules for abstraction according to the circuit diagram are: Abstracting the fabric production device as a resistor in a circuit diagram ,capacitance and inductance The analog components, including resistors ,capacitance and inductance The parameter calculation formulas are: , is the resistivity, is the conductor length, is the cross-sectional area of ​​the conductor; , is the dielectric constant, is the plate area, is the plate spacing; , is the magnetic permeability, is the number of coil turns, is the cross-sectional area of ​​the coil, is the coil length; The connection mode between the circuit and the inductor or resistor on the fabric production device is mapped onto the circuit diagram, the connection mode between each component and the power supply is mapped onto the circuit diagram, and the logical relationship between the circuit components on the fabric production device is mapped onto the circuit diagram.

[0022] In the embodiment of the present invention, in S2, the fault classification of the cloth production device is divided into three types: mechanical part fault, circuit part fault and software part fault; Sample data collection for fabric production devices includes: Monitor the circuit components on the fabric production device. Suppose the monitoring time series is , , at every moment The collected circuit component parameter vector is , , that is, collecting 8 parameters, including heating element current, voltage, resistance value, capacitance value, inductance value, nozzle pressure, nozzle temperature and motor speed; In the task of fabric production performed by the printing machine, whether there is a fault is used as a sample label , Indicates a fault. Indicates that there is no fault, and the parameter vector and sample label are used as sample data ; The cloth production device failures are divided into training sets according to whether the sample data is sufficient. and test set , assuming the total number of samples is , the preset sufficient sample number threshold ,Right now , then randomly select samples as training set , , the remaining samples are used as the test set , where the number of training set samples is and the number of test set samples satisfy ,and ,in, ; Divide the fabric production task execution time into several time periods , in each time period Extract a batch of sample data , select part of the sample data as training sample data, and the rest of the sample data as test sample data. According to the fault conditions and corresponding sample labels, a fault condition database is established for the subsequent training of the fault alarm model. Set the time period The total number of sample data drawn from , select the ratio , then the number of training samples , the number of test samples ,in, .

[0023] In an embodiment of the present invention, in S3, a fault alarm model of a fabric production device is modeled, including the following process: Establish a fault alarm model based on convolutional neural network, which includes convolutional network layer, pooling layer and fully connected layer; In the convolutional network layer, convolution kernels are used to extract features from the input sample data. Each convolutional network layer includes several feature maps and outputs the extracted features. Suppose the input sample data is , the convolution kernel is The size is , convolution step size , then the convolutional layer outputs features The calculation formula is: , in, is the convolution kernel size, is the number of input data channels, is the bias term, with an initial value of 0.2. , , are the row, column, and channel indices of the output feature map, respectively. Represents the position index of the element in the convolution kernel and the number of feature maps output by each convolutional network layer The number of feature maps in the previous layer and the number of convolution kernels The relationship is , let the number of feature maps input to the first convolutional network layer be , the number of convolution kernels ， Then the number of feature maps output by the first layer is ; In the pooling layer, the pooling method is used to compress the features extracted by the convolutional network layer, reduce data redundancy, extract features, reduce prediction errors, and improve prediction accuracy. The pooling window size is set to , then the pooling layer output The calculation formula is: , in, The range is , , , They are the row, column, and channel indices of the output feature map, respectively; In the fully connected layer, the ReLU activation function is used to combine the data output by the convolutional network layer and the pooling layer to output the prediction result. Suppose the input of the fully connected layer is , the weight matrix is , the bias vector is , then the fully connected layer outputs The calculation formula is ,in, , Represents the original data.

[0024] In the embodiment of the present invention, in S4, the accuracy determination of the model includes the following process: The training sample data is distributed to the convolution kernel of the convolutional neural network according to the preset ratio. Suppose the total number of training sample data is , let the number of convolution kernels , then assign to The number of samples of the convolution kernel is ,but ; Set the convolutional neural network structure and set the hyperparameters, including the number of iterations , error function , the error function uses the cross entropy loss function ,in, is the true label of the sample, Predict output for the model; The training sample data is input into the convolutional neural network for training to obtain the convolutional neural network training results. During the training process, the model parameters are updated through the back propagation algorithm. The weight update amount is set and bias update The calculation formulas are , ,in, is the learning rate, ; Input the test sample data into the training results to obtain the predicted data, and compare the predicted data with the labels corresponding to the sample data; Setting the accuracy threshold ,The accuracy threshold,value is 80%, and the accuracy of the fault alarm model is determined based on the comparison results. , let the number of samples predicted correctly be , then the accuracy , then execute S5. If , execute S2.

[0025] In an embodiment of the present invention, in S6, fault identification is performed based on the fault alarm model, including the following process: The collected data of the fabric production device is transmitted as input to the fault alarm model, feature extraction is performed based on the convolutional network layer in the fault alarm model, feature compression is performed based on the pooling layer in the fault alarm model, and the prediction result is output after fusion based on the fully connected layer in the fault alarm model.

[0026] In the embodiment of the present invention, in S7, analyzing and judging whether the device has a fault includes the following process: Analyze the characteristic information and data relationships of the sample data, including determining the number of data collected by the device , determine the frequency of sample data collection , fault identification is performed through the fault alarm model, and the fault probability output by the model is assumed to be ,but Calculated by the fault alarm model in step S3; According to the data distribution and model recognition effect, the probability value of the model recognition result is determined, and this value is set as the recognition threshold of the model. , let the mean of the data distribution be , the standard deviation is , the coefficient is determined based on experience ,but ; Perform input data preprocessing for the fault alarm model, including data normalization and scaling. Suppose the original data is The normalized data is , then the normalization formula is ,set up , ,but , the scaling formula is ,in, is the scaling factor, is the offset, then ; Input the preprocessed data into the fault alarm model, including setting the input layer, hidden layer and output layer of the model and determining the input data and output data. The input data is the preprocessed sample data, and the output data is the fault prediction probability. The recognition result of the data is calculated by the fault alarm model, and the recognition probability is With recognition threshold For comparison, because , so it is considered that there is no fault, otherwise there is a fault; Determine whether there is a fault. If there is no fault, go to step S5; if there is a fault, go to step S8.

[0027] Embodiment 2, as Figure 2 As shown, a fault warning system for fabric production equipment includes: A device parameter identification module, used for establishing parameters required to set a fault alarm model according to a predetermined fabric production device; A sample collection module, used to collect sample data according to the parameters determined by the device parameter identification module; A model building module is used to build a fault alarm model based on the collected sample data; The accuracy judgment module is used to judge whether the accuracy of the model meets the requirements according to the established fault alarm model; A real-time data acquisition module is used to collect data of the material distribution device in real time according to the established fault alarm model; A fault identification module, used for performing fault identification based on the collected sample data; The alarm module is used to generate an alarm based on the identification result of the fault identification module.

[0028] The embodiments of the present invention disclose preferred embodiments, but are not limited thereto. A person skilled in the art can easily understand the spirit of the present invention based on the above embodiments and make different extensions and changes. However, as long as they do not deviate from the spirit of the present invention, they are all within the protection scope of the present invention.

Claims

1. A method for early warning of failure of fabric production equipment, characterized in that: The following steps are involved: S1. Parameter determination; Determine parameters based on a predetermined fabric production device and parameters required to establish a fault alarm model; S2, fault classification and sample collection; Based on the parameters determined in step S1, fault classification of the fabric production device and corresponding sample data collection; S3, fault alarm model building; Based on the sample data collected in step S2, a fault alarm model is built; S4, model accuracy judgment; Based on the fault alarm model established in step S3, the accuracy of the fault alarm model is determined and whether the model accuracy meets the requirements is judged. If it does not meet the requirements, step S2 is executed; if it meets the requirements, step S5 is executed; S5, device data collection; Based on the fault alarm model established in step S3, data of the fabric production device is collected; Determine whether the data collected from the current fabric production device is sufficient, if not, collect new sample data, otherwise, proceed to step S6; The collected data of the fabric production device is sent to the fault alarm model for identification. If the accuracy of the identification result is greater than the set accuracy threshold , then proceed to step S7, otherwise, further collect sample data according to step S2. S6, fault identification; Based on the fault alarm model established in step S3, fault identification is performed on the collected data of the fabric production device; S7. Fault analysis and judgment; Based on the recognition result of step S6, whether there is a fault in the fabric production device is analyzed and judged, if there is no fault, step S5 is executed, if there is a fault, step S8 is executed; S8, alarm is issued; Based on the judgment result of step S7, an alarm instruction is sent to the alarm device via the signal calling device.

2. A method for early warning of a fault of a fabric production equipment according to claim 1, characterized in that: In S1, the parameters of the fabric production device are determined The process includes: The fabric production device is abstracted into a physical diagram according to the device parameters; Establish the relationship between the physical diagram and the circuit diagram, and convert the physical diagram of the fabric production device into a circuit diagram; Establish the equivalent transformation between the physical diagram and the circuit diagram, and the rules for abstraction according to the circuit diagram are: Abstracting the fabric production device as a resistor in a circuit diagram ,capacitance and inductance The analog components, including resistors ,capacitance and inductance The parameter calculation formulas are: , is the resistivity, is the conductor length, is the cross-sectional area of ​​the conductor; , is the dielectric constant, is the plate area, is the plate spacing; , is the magnetic permeability, is the number of coil turns, is the cross-sectional area of ​​the coil, is the coil length; The connection mode between the circuit and the inductor or resistor on the fabric production device is mapped onto the circuit diagram, the connection mode between each component and the power supply is mapped onto the circuit diagram, and the logical relationship between the circuit components on the fabric production device is mapped onto the circuit diagram.

3. A method for early warning of a fault of a fabric production device according to claim 2, characterized in that: In S2, the fault classification of the fabric production device is divided into three types: mechanical part fault, circuit part fault and software part fault; Sample data collection for fabric production devices includes: Monitor the circuit components on the fabric production device. Suppose the monitoring time series is , at every moment The collected circuit component parameter vector is ; In the fabric production task performed by the fabric production device, whether there is a fault is used as a sample label , Indicates a fault. Indicates that there is no fault, and the parameter vector and sample label are used as sample data ; The cloth production device failures are divided into training sets according to whether the sample data is sufficient. and test set , assuming the total number of samples is ,like , then randomly select samples as training set , the remaining samples are used as the test set , where the number of training set samples is and the number of test set samples satisfy ,and ,in, is the preset sufficient sample number threshold, ; Divide the fabric production task execution time into several time periods , in each time period Extract a batch of sample data , select part of the sample data as training sample data, and the rest of the sample data as test sample data. According to the fault conditions and corresponding sample labels, a fault condition database is established for the subsequent training of the fault alarm model. Set the time period The total number of sample data drawn from , select the ratio , then the number of training samples , the number of test samples ,in, .

4. A method for early warning of a fault of a fabric production device according to claim 3, characterized in that: In S3, a fault alarm model of a fabric production device is modeled. The process includes: Establish a fault alarm model based on convolutional neural network, which includes convolutional network layer, pooling layer and fully connected layer; In the convolutional network layer, convolution kernels are used to extract features from the input sample data. Each convolutional network layer includes several feature maps and outputs the extracted features. Suppose the input sample data is , the convolution kernel is , the convolution step size is , then the convolutional layer outputs features The calculation formula is: ， in, is the convolution kernel size, is the number of input data channels, is the bias term, , , are the row, column, and channel indices of the output feature map, respectively. Represents the position index of the element in the convolution kernel and the number of feature maps output by each convolutional network layer The number of feature maps in the previous layer and the number of convolution kernels The relationship is ; In the pooling layer, the pooling method is used to compress the features extracted by the convolutional network layer, reduce data redundancy, extract features, reduce prediction errors, and improve prediction accuracy. The pooling window size is set to , then the pooling layer output The calculation formula is: ， in, The range is , , , They are the row, column, and channel indices of the output feature map, respectively; In the fully connected layer, the ReLU activation function is used to combine the data output by the convolutional network layer and the pooling layer to output the prediction result. Suppose the input of the fully connected layer is , the weight matrix is , the bias vector is , then the fully connected layer outputs The calculation formula is ,in, , Represents the original data.

5. A method for early warning of a fault of a fabric production device according to claim 4, characterized in that: In S4, the accuracy judgment of the model includes the following process: The training sample data is distributed to the convolution kernel of the convolutional neural network according to the preset ratio. Suppose the total number of training sample data is , assigned to The number of samples of the convolution kernel is ,but ,in, is the number of convolution kernels; Set the convolutional neural network structure and set the hyperparameters, including the number of iterations , error function , the error function uses the cross entropy loss function ,in, is the true label of the sample, Predict output for the model; The training sample data is input into the convolutional neural network for training to obtain the convolutional neural network training results. During the training process, the model parameters are updated through the back propagation algorithm. The weight update amount is set and bias update The calculation formulas are , ,in, is the learning rate; Input the test sample data into the training results to obtain the predicted data, and compare the predicted data with the labels corresponding to the sample data; Setting the accuracy threshold The accuracy threshold is 80%, and the accuracy of the fault alarm model is determined based on the comparison results. , let the number of samples predicted correctly be , then the accuracy ,like , execute S5, if , execute S2.

6. A method for early warning of a fault of a fabric production device according to claim 5, characterized in that: In S6, fault identification is performed based on the fault alarm model, including the following process: The collected data of the fabric production device is transmitted as input to the fault alarm model, feature extraction is performed based on the convolutional network layer in the fault alarm model, feature compression is performed based on the pooling layer in the fault alarm model, and the prediction result is output after fusion based on the fully connected layer in the fault alarm model.

7. A method for early warning of a fault of a fabric production equipment according to claim 1, characterized in that: In S7, analyzing and judging whether the device has a fault includes the following process: Analyze the characteristic information and data relationships of the sample data, including determining the number of data collected by the device , determine the frequency of sample data collection , fault identification is performed through the fault alarm model, and the fault probability output by the model is assumed to be ,but Calculated by the fault alarm model in step S3; According to the data distribution and model recognition effect, the probability value of the model recognition result is determined, and this value is set as the recognition threshold of the model. , let the mean of the data distribution be , the standard deviation is ,but ,in, is a coefficient determined empirically; Perform input data preprocessing for the fault alarm model, including data normalization and scaling. Suppose the original data is The normalized data is , then the normalization formula is , the scaling formula is ,in, is the scaling factor, is the offset; Input the preprocessed data into the fault alarm model, including setting the input layer, hidden layer and output layer of the model and determining the input data and output data. The input data is the preprocessed sample data, and the output data is the fault prediction probability. The recognition results of the data calculated by the fault alarm model include the recognition probability of the set output , comparing the recognition probability with the recognition threshold, if the recognition probability is greater than the threshold, it is considered that there is a fault, otherwise there is no fault; Determine whether there is a fault. If there is no fault, go to step S5; if there is a fault, go to step S8.

8. A system applied to the method for early warning of a fault of a fabric production device as claimed in claim 7, characterized in that: include: A device parameter identification module, used for establishing parameters required to set a fault alarm model according to a predetermined fabric production device; A sample collection module, used to collect sample data according to the parameters determined by the device parameter identification module; A model building module is used to build a fault alarm model based on the collected sample data; The accuracy judgment module is used to judge whether the model accuracy meets the requirements according to the established fault alarm model; A real-time data acquisition module is used to collect data of the material distribution device in real time according to the established fault alarm model; A fault identification module, used for performing fault identification based on the collected sample data; The alarm module is used to generate an alarm based on the identification result of the fault identification module.

Images