Solenoid valve abnormity detection method and system of jacquard machine

Abstract

The invention relates to the technical field of electromagnetic valve detection, in particular to a jacquard electromagnetic valve anomaly detection method and system, and the method comprises the steps: collecting and recording a control signal, a current signal and a magnetic signal of an electromagnetic valve; based on a deep learning technology, constructing a model fusing a convolutional neural network and a long-short term memory network, and measuring response time; on the basis of a generative adversarial network, current stability is measured according to current signal data collected by a high-precision current sensor in combination with environmental parameters; a quantum enhanced sensor is adopted, a quantum mechanical model is constructed by considering various environmental parameters through a quantum algorithm, a measured original magnetic signal is calibrated and then compared with a standard value, and the magnetic intensity of the electromagnetic valve is measured; and constructing a fault diagnosis model based on a fault tree, determining a logic relationship between events according to historical fault data, analyzing the probability and logic of the events, and positioning the fault cause of the electromagnetic valve. The method comprehensively considers the influence of environmental factors, and improves the detection efficiency and comprehensiveness.

Description

Technical Field

[0001] The present invention relates to the technical field of solenoid valve detection, and particularly relates to a method and system for detecting abnormalities of solenoid valves in a jacquard machine. Background Art

[0002] In the textile industry, as the core equipment for producing delicate and complex fabrics, the advanced level of the technology of a jacquard machine directly determines the quality and production efficiency of the fabrics. Solenoid valves play a crucial role in a jacquard machine. It precisely controls the interweaving actions of warp and weft threads, and through electromagnetic force, it precisely controls components such as knitting needles and selector devices, thereby weaving various complex and delicate pattern designs. When the jacquard machine is working, the solenoid valves act frequently, and extremely high requirements are imposed on their response speed, stability, and control accuracy. For example, when producing high-end silk jacquard fabrics, the solenoid valves need to precisely control the up and down movement of the knitting needles within a short period of time. Any deviation in each action may result in pattern defects, seriously affecting the product quality. However, there are many drawbacks in the traditional detection methods for solenoid valves in a jacquard machine. The traditional single-channel detection method can only monitor a certain parameter of the solenoid valve and cannot comprehensively obtain its operating state information. Taking the detection of current as an example, only focusing on the current parameter, it is impossible to know whether the control signal is stable. Once the control signal is abnormal, it may cause the solenoid valve to malfunction, and then the jacquard pattern will be disordered, but this potential fault is difficult to be detected in time. Moreover, the working environment of the jacquard machine is complex. The humidity in the workshop is relatively high, the static electricity generated during the processing of silk threads, and the electromagnetic interference generated by surrounding electrical equipment, etc., will all have a significant impact on the performance of the solenoid valves. A high-humidity environment may cause the internal components of the solenoid valve to rust, increasing the movement resistance and affecting the response speed; strong electromagnetic interference may interfere with the transmission of the control signal, making the actions of the solenoid valves unable to be precisely executed. However, the traditional detection technology is unable to cope with these complex environmental factors, and it is difficult to accurately detect the response time and current stability of the solenoid valves, and it is unable to effectively identify the potential fault hazards caused by environmental interference. In the fault diagnosis link, the traditional method mainly relies on manual experience judgment. When a solenoid valve of a jacquard machine fails, technicians rely on their own experience to troubleshoot possible causes of the failure. This method is highly subjective and inefficient. Due to the complex structure of the jacquard machine and numerous possible causes of faults related to the solenoid valves, manual troubleshooting often takes a lot of time and may repeatedly check in multiple unnecessary links, resulting in a significant extension of the downtime of the jacquard machine. This not only seriously affects the production progress, increases the production cost, but also may have a negative impact on the enterprise's reputation due to delivery delays. Therefore, to ensure the stable and efficient operation of the jacquard machine and produce high-quality fabrics, there is an urgent need to develop a comprehensive and accurate detection method and an efficient fault diagnosis scheme specifically for the solenoid valves of the jacquard machine to meet the growing production needs of the textile industry. Summary of the Invention

[0003] The present invention comprehensively considers the influence of environmental factors, improves the fault detection accuracy of the jacquard machine solenoid valve, and ensures the stable and efficient operation of the jacquard machine.

[0004] The technical solution proposed by the present invention is: a method for detecting abnormalities of the solenoid valve of a jacquard machine, the method comprising: Adopt multi-channel parallel detection, collect and record the control signal, current signal and magnetic force signal of the solenoid valve; Based on deep learning technology, construct a model that combines a convolutional neural network and a long short-term memory network, integrate environmental parameters to perform feature extraction and time series analysis on the solenoid valve control signal, and measure the response time; Based on the generative adversarial network, by processing the extended input obtained by splicing the input noise and the parametric linear transformation of the environmental parameters, the generator and the discriminator are trained adversarially, and based on the current signal data collected by the high-precision current sensor and combined with the environmental parameters, the current stability is measured; Adopt a quantum-enhanced sensor, construct a quantum mechanics model by considering various environmental parameters through a quantum algorithm, calibrate the measured original magnetic force signal and compare it with the standard value to measure the magnetic force intensity of the solenoid valve; Construct a fault diagnosis model based on a fault tree, determine the logical relationship between events according to historical fault data, analyze the event probability and logic, and locate the cause of the solenoid valve fault.

[0005] Preferably, the construction of the model that combines a convolutional neural network and a long short-term memory network includes the following steps: ; Wherein: is the input signal, is the dimension of the original input signal, is the temperature, is the humidity, is the electromagnetic interference intensity, is the dust concentration, is the chemical agent concentration, is the extended input, the output of the th convolutional layer is calculated by the following formula: ; Wherein: is the weight of the th convolutional layer and the th convolutional kernel, is the th channel of the extended input , is the bias of the th convolutional layer, represents the convolution operation; Integrate environmental parameters into the LSTM model, and the environmental parameters are integrated through the following transformation: ; Where: and are the mean and standard deviation of temperature in the training set, respectively, and are the mean and standard deviation of humidity in the training set, respectively, and are the mean and standard deviation of electromagnetic interference intensity in the training set, respectively, and are the mean and standard deviation of dust concentration in the training set, respectively, and are the mean and standard deviation of chemical agent concentration in the training set, respectively, is the input signal input into the LSTM model, is the extended input that adds environmental factors to .

[0006] Preferably, the training process of the model integrating the convolutional neural network and the long short-term memory network is as follows: Preset the exponential decay rates of the learning rate, the first moment, and the second moment; in each round of training, calculate the gradients of the model parameters according to the loss function; the Adam optimizer dynamically adjusts the model parameters according to these gradients; regularly evaluate the performance of the model on the validation set during training. If the value of the loss function does not decrease after several consecutive iterations, appropriately reduce the learning rate to prevent the model from falling into a local optimum; continuously loop through the training until the model reaches satisfactory accuracy and error metrics on the validation set.

[0007] Preferably, the detection process of the current stability is as follows: Concatenate the input noise vector with the linearly transformed environmental parameters to form an extended input; calculate the generator output through the weight matrices and bias vectors of different layers, combined with activation functions. The ReLU activation function is used for the hidden layer, and the Tanh activation function is used for the output layer; use the current data collected by the high-precision current sensor combined with the environmental parameters as the input, and optimize the performance of the generator and discriminator through adversarial training. The cross-entropy loss function is used to measure the training effect.

[0008] Preferably, the calculation formula for the magnetic force intensity of the solenoid valve is as follows: ; Where: is the original magnetic force intensity, is the calibrated magnetic force intensity, is the temperature, is the humidity, is the electromagnetic interference intensity, is the dust concentration, is the chemical agent concentration.

[0009] Preferably, the specific content of the fault diagnosis model is as follows: Taking the solenoid valve fault as the top event, setting the coil fault, spool fault, control circuit fault, and environmental factors as intermediate and lower events; analyzing the historical fault data to clarify the logical relationship between each event; combining the specific fault phenomena and inferring the fault cause based on the determined logical relationship; continuously collecting new fault cases to update the event occurrence probability and logical relationship.

[0010] Preferably, each channel of the multi-channel parallel detection is mutually shielded, and a filtering algorithm is added to the signal processing current. A high-speed voltage sensor is used to collect the control signal; a high-precision current sensor is used to collect the current signal; a quantum-enhanced magnetic sensor is used to collect the magnetic force signal.

[0011] The present invention also provides a solenoid valve abnormality detection system for a jacquard machine, and the system is used to execute the solenoid valve abnormality detection method for a jacquard machine described above.

[0012] The present invention also provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement the solenoid valve abnormality detection method for a jacquard machine described above.

[0013] Advantages of the present invention: By using the multi-channel parallel detection means, it can synchronously collect control signals, current signals, magnetic force signals, etc. Sensors adapted to different parameter detections in different channels perform their respective functions and process data in parallel. This not only greatly shortens the detection time, but also can obtain multi-parameter operation data at one time, comprehensively reflecting the operation state of the solenoid valve, changing the limitations of the previous single-parameter detection, greatly improving the detection efficiency and comprehensiveness, and enabling the operator to quickly master the overall condition of the solenoid valve.

[0014] A response time detection model based on deep learning incorporates environmental parameters such as temperature, humidity, and electromagnetic interference intensity. The convolutional neural network and the long short-term memory network work together to effectively extract the characteristics of the solenoid valve response signal in a complex environment, analyze the time series relationship, and accurately detect the response time. When the generative adversarial network detects the current stability, the input noise vector and environmental parameters are concatenated and input after linear transformation to optimize the model to adapt to environmental interference. The quantum-enhanced sensor adapts to complex environments with a special coating and constructs a model through quantum algorithms to calibrate the magnetic force intensity. These technologies enable the detection method to accurately detect under harsh conditions such as high temperature, high humidity, and strong electromagnetic interference, broadening the application range of solenoid valves under complex working conditions.

[0015] Construct a fault diagnosis model based on a fault tree. Taking the solenoid valve fault as the top event, subdivide the intermediate and bottom events such as coil, spool, control circuit faults, and the influence of environmental factors, and further subdivide the environmental factor sub-events. Determine the logical relationship between events through a large amount of fault data, and update the event probabilities and logic in combination with actual cases to accurately locate the cause of the fault. Compared with traditional fault diagnosis methods, this model is more systematic and comprehensive, can quickly troubleshoot complex faults, reduce maintenance time and costs, and improve the reliability of equipment operation. Brief Description of the Drawings

[0016] Figure 1 It is a flowchart of a method and system for detecting abnormalities of solenoid valves of a jacquard machine according to the present invention; Figure 2 It is a flowchart of a fault diagnosis model of a method and system for detecting abnormalities of solenoid valves of a jacquard machine according to the present invention. Detailed Embodiments

[0017] The following description is used to disclose the present invention so that those skilled in the art can implement the present invention. The preferred embodiments described below are only examples, and those skilled in the art can think of other obvious variations. The basic principles defined in the following description can be applied to other implementation schemes, deformation schemes, improvement schemes, equivalent schemes, and other technical schemes that do not depart from the spirit and scope of the present invention.

[0018] It can be understood that the term "one" should be understood as "at least one" or "one or more". That is, in one embodiment, the number of an element can be one, and in other embodiments, the number of the element can be multiple. The term "one" cannot be understood as a limitation on the number.

[0019] Such as Figure 1As shown in the figure, this solution adopts multi-channel parallel detection technology, and independent detection channels are set for the control signal, current signal, magnetic force signal, etc. of the solenoid valve. Each channel is equipped with a dedicated sensor and signal processing circuit to achieve synchronous detection of different parameters. For example, in the channel for detecting the response time of the solenoid valve, a high-speed voltage sensor is used to collect the control signal; in the channel for detecting current stability, a high-precision current sensor is adopted; in the channel for detecting magnetic force intensity, a quantum-enhanced magnetic sensor is used. The data acquisition and processing of each channel are carried out in parallel, greatly improving the detection efficiency.

[0020] During the operation of the jacquard machine, the multi-channel parallel detection system is started. The channel for detecting the response time collects the control signal at a sampling frequency of 10 kHz, the channel for detecting current stability collects the current signal at a sampling frequency of 5 kHz, and the channel for detecting magnetic force intensity collects the magnetic force signal at a sampling frequency of 2 kHz. Through parallel processing, multi-parameter data such as the response time, current stability, and magnetic force intensity of the solenoid valve at this moment can be obtained within 1 second. At the same time, in order to avoid signal crosstalk between channels, shielded cables are used to transmit signals, and a filtering algorithm is added to the signal processing circuit. For example, for the electromagnetic interference signals in the channel for detecting current stability, the interference in a specific frequency range is filtered out through a band-pass filter to ensure the accuracy and reliability of the collected current signal. Through multi-channel parallel detection, the operating parameters of the solenoid valve in a complex environment can be quickly and comprehensively obtained, providing rich data support for subsequent fault diagnosis and performance evaluation.

[0021] Deep learning technology is adopted to construct a model based on the fusion of convolutional neural network (CNN) and long short-term memory network (LSTM) for response time detection.

[0022] Let the input signal be , the dimension of the original input signal is , the environmental parameters temperature , humidity , electromagnetic interference intensity , dust concentration , chemical agent concentration are taken as one dimension respectively. The environmental parameters are incorporated into the input as additional channel information to form the extended input . The construction process of the extended input is as follows: ; At this time, the dimension of the extended input becomes . The output of the th convolutional layer is calculated by the following formula: ; Wherein: is the weight of the th convolutional kernel in the th convolutional layer, is the th channel of the extended input and is the bias of the th convolutional layer, represents the convolution operation. Here, three convolutional layers are set, with the convolutional kernel sizes being 3×3, 5×5, and 3×3 respectively, and the stride being 1 for all. Through these convolutional layers, feature extraction is performed on the control signal containing environmental parameters and the solenoid valve action signal to capture the local features in the signal.

[0023] For example, assume the original input signal is the voltage change sequence of the solenoid valve control signal, with a dimension of . At a certain moment, the temperature , the humidity , the electromagnetic interference intensity (in the unit of a certain electromagnetic interference intensity index value), the dust concentration , and the chemical agent concentration . These environmental parameters are combined with the original input signal to form the extended input , whose dimension becomes 100 + 5 = 105. When performing operations in the first convolutional layer (with a convolutional kernel size of 3×3), the convolutional kernel performs a convolution operation with . For example, when , performs a convolution operation on the first 3 elements of the original input signal and environmental parameters such as the temperature . The output result is then processed by the ReLU activation function to obtain the output value at the corresponding position of this convolutional layer. In a high-temperature environment, the weight of the convolutional kernel will be adjusted during training to adapt to the signal characteristic changes caused by temperature. For example, when the environmental temperature rises, some weights related to temperature-sensitive features will increase to highlight the extraction of these features. And when calculating , the numerical change of the temperature will directly affect the input features and thus the convolution operation result. As the temperature rises, assume the weight related to temperature-sensitive features is adjusted from the initial 0.5 to 0.7. This enhances the extraction strength of temperature-related features in the convolution operation, making the output result of this convolutional layer better reflect the signal changes in a high-temperature environment To incorporate environmental parameters into the LSTM model, at the input gate , Forget Gate , Output Gate and Memory Cell in the calculation formula, the input is extended by adding environmental parameter information to form . Let the original input have a dimension of , and the environmental parameters are incorporated through the following transformation: ; where: , are the mean and standard deviation of the temperature in the training set, , are the mean and standard deviation of the humidity in the training set, , are the mean and standard deviation of the electromagnetic interference intensity in the training set, , are the mean and standard deviation of the dust concentration in the training set, , are the mean and standard deviation of the chemical agent concentration in the training set. This is done to normalize environmental parameters with different dimensions to make the model training more stable. At this time, the dimension of becomes +5.

[0024] ; ; ; ; ; ; where: is the input at the current moment containing environmental parameters, is the hidden state at the previous moment, is the weight matrix, is the bias vector, is the sigmoid activation function, and tanh is the hyperbolic tangent activation function, is element-wise multiplication. The LSTM is set with 2 hidden layers, and each hidden layer contains 128 neurons, which are used to learn the long-term dependencies of signals containing the influence of environmental factors in the time series.

[0025] For example, assume the original input is the solenoid valve control signal sequence at a certain moment, with dimension . The mean value of temperature and standard deviation in the training set; The mean value of humidity and standard deviation , etc. When the temperature , humidity , electromagnetic interference intensity , dust concentration , and chemical agent concentration at a certain moment, the environmental parameters are normalized to , and combined with the original input to form , and the dimension becomes 50 + 5 = 55. When calculating the input gate , is input into the formula, where is the weight matrix of the input gate, is the hidden state at the previous moment. For example, a certain row of is multiplied by the corresponding elements of and accumulated, and then the bias is added. Finally, the value of is obtained through the sigmoid function for subsequent calculations of the memory unit and hidden state. In an environment with strong electromagnetic interference, the weight matrix of the LSTM will be adjusted during training so that the model has better processing ability for the time series characteristics of interference signals. For example, when the radio frequency interference intensity increases, the weights related to the time characteristics of the interference signal, etc. will be adjusted. At the same time, the change of environmental parameters (such as the increase of electromagnetic interference intensity ) will affect the calculation process of the LSTM unit through , enhancing the filtering of interference signals and the memory of real signals. As the electromagnetic interference intensity increases, if the weight related to the time characteristics of the interference signal in was 0.3 before and becomes 0.4 after adjustment, this increases the consideration of the characteristics related to the interference signal when calculating , , and , thus enabling the LSTM to better handle electromagnetic interference when processing signals. The adjusted weights will affect the subsequent

[0026] During the model training process, to cope with complex environments, a large number of signal samples under different environmental conditions (such as high temperature, high humidity, electromagnetic interference, dust pollution, etc.) are added. The Adam optimizer is used, and the formula for updating parameters is ; ; ; ; ; where: is the loss function, is the gradient of the loss function with respect to the parameter ; , are the exponential decay rates of the first moment and the second moment respectively (usually set = 0.9, = 0.999), is the learning rate set to 0.001, is the number of un-iterated times, is a small constant to prevent the denominator from being zero, and the number of training rounds is 200. During the training process, the model performance is regularly evaluated on the validation set, and the mean squared error (MSE) is used as the evaluation metric. The calculation formula is , where is the true response time, is the response time predicted by the model, is the number of samples in the validation set, represents the first moment estimate, represents the second moment estimate, represents the bias correction for , represents the bias correction for , represents the model parameters updated at the moment, represents . According to the change of MSE on the validation set, if the MSE does not decrease for 5 consecutive iterations, the learning rate is appropriately reduced (such as reduced to 0.8 times the original) to avoid the model falling into a local optimum.

[0027] For example, at the beginning of training, when the model processes a batch of samples from a high-temperature and high-humidity environment, since the response time of the solenoid valve may become longer in a high-temperature environment, there is a large error between the response time predicted by the model and the true response time , resulting in a large value of the loss function . The parameters are updated through the Adam optimizer. At the = 10, it is calculated that and , and then calculate and , update the parameters according to the formula . As the training progresses, samples under different environmental conditions contribute differently to the update of the parameters . For example, in the training stage with more samples in a high-humidity environment, the parameters corresponding to the signal features related to humidity will have a relatively larger update amplitude. Humidity, as a part of, its numerical change will affect the calculation of the loss function, and thus affect the update of the parameters , so that the model gradually adapts to the change characteristics of signals in a high-humidity environment. On the validation set, after 100 iterations, the MSE value starts to show the situation of not decreasing for 5 consecutive times. At this time, the learning rate is reduced to 0.0008, and continue to train to optimize the model performance. In the training stage with more samples in a high-humidity environment, assume that the proportion of the parameters corresponding to humidity in the loss function calculation is 0.2. As the training progresses, if the high-humidity samples increase this part of the contribution in the loss function, resulting in an overall increase in the loss function value, when updating the parameters through the Adam optimizer, the update amplitude of the parameters related to humidity will increase accordingly. For example, the original update amount is 0.005, and at this time it may increase to 0.008, so that the model can better process signals in a high-humidity environment in the subsequent training. As the training continues, the model gradually deepens its learning of signal features in different environments. When the samples of electromagnetic interference environment increase, the adjustment of the parameters related to electromagnetic interference becomes more significant. For example, in a certain training, the numerical value of the electromagnetic interference intensity fluctuates frequently, resulting in the weights and corresponding to the time features of the interference signal in the LSTM are continuously updated. After 50 iterations, the weights related to the time features of the interference signal in increase from 0.4 to 0.5 further, which makes the model significantly enhance its ability to identify and process interference signals when calculating the input gate and the forget gate . When facing new samples of electromagnetic interference environment subsequently, the model can more accurately filter out interference signals and capture the true features of the solenoid valve control signal, thereby improving the accuracy of predicting the response time of the solenoid valve. When the accuracy rate of the model on the validation set reaches 80% and the MSE is stable within 0.05, it is considered that the detection performance of the model for the response time of the solenoid valve in a complex environment reaches the expectation and can be used for actual detection tasks.

[0028] Using the Generative Adversarial Network (GAN) technology, the generator adopts a fully connected neural network structure, including 3 hidden layers, each with 256 neurons, and the number of neurons in the output layer is determined according to the dimension of the current data. The discriminator also uses a fully connected neural network with 2 hidden layers, each with 128 neurons.

[0029] Let the input noise vector be , the original noise vector with a dimension of , the environmental parameter temperature , humidity , electromagnetic interference intensity , dust concentration After the following linear transformation, they are concatenated with the noise vector to form an extended input : ; where: , , , are linear transformation coefficients, , , , are bias terms, and these coefficients and bias terms are adjusted through the backpropagation algorithm during the training process to optimize the model's ability to model the relationship between environmental parameters and current data. At this time, the dimension of the extended input becomes +4. The output of the th layer of the generator is: ; where: is the weight matrix of the th layer, is the bias vector, is the activation function.

[0030] For example, assume that the dimension of the input noise vector =10. At the initial stage of training, the linear transformation coefficient =0.1, =0; =0.1, =0, etc. When the temperature =40°C, the humidity =75%, the electromagnetic interference intensity =45, and the dust concentration =0.04 When it is, after linear transformation, we get [0.1×40 + 0, 0.1×75 + 0, 0.1×45 + 0, 0.1×0.04 + 0], which is concatenated with the noise vector to form the extended input , and the dimension becomes 10 + 4 = 14. When operating on the first hidden layer (256 neurons) of the generator, the weight matrix is multiplied with (i.e., ) in a matrix multiplication operation. For example, a certain row of is multiplied with the elements of and accumulated, then added with the bias , and finally the output of this layer is obtained through the ReLU activation function . As the training progresses, when the number of samples in the high-temperature environment increases, the linear transformation coefficient may be adjusted to 0.15, which increases the influence of temperature on the output of the generator, enabling the generator to better simulate the characteristics of current data in the high-temperature environment.

[0031] Suppose the input data (real current data or data generated by the generator) is , and the dimension of the original input data is . The environmental parameters (temperature , humidity , electromagnetic interference intensity , dust concentration ) are linearly transformed as follows and concatenated with the input data to form the extended input .

[0032] ; where: , , , are the linear transformation coefficients, and , , , are the bias terms, which are also optimized through the backpropagation algorithm during training. At this time, the dimension of the extended input becomes +4. The output of the th layer of the discriminator is: ; where: is the weight matrix of the th layer, is the bias vector, It is an activation function (here, the ReLU activation function is used for the hidden layer, and the Sigmoid activation function is used for the output layer).

[0033] For example, assume the input real current data has a dimension = 20. During the training process, when the electromagnetic interference intensity varies greatly, the linear transformation coefficients of the discriminator will be adjusted accordingly. For example, initially = 0.08, = 0. As the number of electromagnetic interference environment samples increases, it is adjusted to = 0.12. When an electromagnetic interference-affected current data is input, and the environmental parameters are = 38°C, = 72%, = 50, = 0.035 at that time, after linear transformation, it is concatenated with to form an extended input , and the dimension becomes 20 + 4 - 24. When operating in the first hidden layer (128 neurons) of the discriminator, the weight matrix and (i.e., ) perform a matrix multiplication operation, and after being processed by the ReLU activation function, it obtains . In this way, the discriminator can more accurately identify the current data affected by electromagnetic interference and distinguish between real data and the data generated by the generator.

[0034] Take the current data collected by the high-precision current sensor as the input, and at the same time combine the environmental parameters (temperature , humidity , electromagnetic interference intensity , dust concentration ) as auxiliary inputs. After the environmental parameters are linearly transformed, they are concatenated with the current data and input into the model. During the training process, the performance of the generator and the discriminator is continuously optimized through adversarial training. The cross-entropy loss function is used during training, and the loss functions of the generator and the discriminator are respectively:[[]] ; ; Among them: is the extended noise distribution, is the extended real data distribution. The learning rates of both the generator and the discriminator are set to 0.0002, the training batch size is 64, and the number of training epochs is 300 times.

[0035] For example, in the 100th round of training, the difference between the data generated by the generator and the real current data is relatively large. The discriminator can accurately identify most of the generated data, resulting in a higher value of the loss function of the generator. Through the backpropagation algorithm, the generator adjusts the linear transformation coefficients and weight matrices. For example, is adjusted from 0.15 to 0.13 to better simulate the distribution of real current data. At the same time, the discriminator also adjusts its own linear transformation coefficients and weight matrices according to the new generated data and real data to improve its discrimination ability for data. During the training process, by monitoring the loss function values of the generator and the discriminator and the similarity between the generated data and the real data (such as measured by the mean square error), when the mean square error between the data generated by the generator and the real data is less than 0.08 and the accuracy of the discriminator reaches 75%, it is considered that the model training has achieved good results and can be used to detect the current stability of the solenoid valve. In the training stage with a large number of samples in a high-humidity environment, the linear transformation coefficients and weight matrices of the generator and the discriminator will be adjusted according to the humidity factor, so that the model can accurately simulate and discriminate the current data in a high-humidity environment, ensuring the accuracy of detecting the current stability of the solenoid valve in a complex environment.

[0036] Utilize a quantum-enhanced magnetic sensor, which has ultra-high sensitivity and strong anti-interference ability and can adapt to complex environments. The surface of the sensor is coated with a special moisture-proof, heat-dissipating, and dust-proof coating to ensure stable operation in high-temperature, high-humidity, and multi-dust environments. Process the measured data through a quantum algorithm, fully considering the comprehensive influence of environmental parameters (temperature , humidity , electromagnetic interference intensity , dust concentration , chemical agent concentration ) on magnetic force measurement, and establish a complex quantum mechanics model. Let the measured original magnetic force intensity be , and the calibrated magnetic force intensity The calculation formula is: ; Compare the calibrated magnetic force intensity with the standard value to accurately judge whether the magnetic force of the solenoid valve is normal.

[0037] For example, in a certain detection scenario, the quantum-enhanced magnetic sensor measures the original magnetic force intensity = 0.8 , and the environmental parameters at this time are = 42°C, = 78%, = 48, = 0.045 , = 0.09 Substitute these environmental parameters into the calibration formula, and the calculated result is 0.3×42 + 0.2×78 + 0.25×48 + 0.15×0.045 + 0.1×0.09 = 12.6 + 15.6 + 12 + 0.00675 + 0.009 = 40.21575. Then the calibrated magnetic force intensity = 0.8 - 0.04021575 = 0.75978425 If the standard magnetic force intensity of this type of solenoid valve is 0.75 - 0.85 then it can be judged that the magnetic force of the solenoid valve is normal at this time. In practical applications, statistical analysis is carried out on a large number of measurement data of different types of solenoid valves, and the error distribution between the calibrated magnetic force intensity and the standard value is calculated. If the proportion of samples with errors exceeding a certain range (such as ±0.05T) exceeds 5%, then re-evaluate the rationality of the environmental parameter weight coefficients in the quantum algorithm, and adjust the weight coefficients through methods such as fitting experimental data to improve the calibration accuracy. As the environmental parameters change, the quantum algorithm will adjust the calibration of the magnetic force intensity in real time to ensure accurate detection of the magnetic force intensity of the solenoid valve in complex environments.

[0038] Such as Figure 2 shown, construct a fault diagnosis model based on a fault tree, with the solenoid valve fault (denoted as ) as the top event. Set various factors that may cause faults, such as coil fault (denoted as ), spool fault (denoted as ), control circuit fault (denoted as ), environmental factor influence (denoted as ) etc. as intermediate events, while too high temperature (denoted as ), too high humidity (denoted as ), excessive electromagnetic interference (denoted as ), serious dust pollution (denoted as ), chemical agent corrosion (denoted as ) etc. are used as bottom events, which belong to the intermediate event of environmental factor influence.

[0039] Through in-depth analysis of a large amount of fault data, determine the logical relationship between each event. Taking the AND gate logic as an example, assume that when the temperature is too high ( > 45°C) and the electromagnetic interference exceeds the standard ( > 50) occur simultaneously, it may cause a control circuit fault. Let the probability of the event of too high temperature occurring be , and the probability of the event of excessive electromagnetic interference occurring be . Based on the AND gate logic, the probability that these two conditions are simultaneously satisfied and cause a control circuit fault It can be calculated by the following formula: ; wherein, is the correlation coefficient considering the combined effect of over - high temperature and excessive electromagnetic interference leading to control circuit failure, and its value range is 0 - 1. This coefficient is obtained through statistical analysis of the relationship between these two factors and control circuit failure in a large number of actual failure cases. For example, after studying 1000 relevant failure cases, it is found that 300 cases have control circuit failure under the condition of over - high temperature and excessive electromagnetic interference. Then . If in a certain working condition, through monitoring (i.e., the probability of the over - high temperature event occurring is 20%), (i.e., the probability of the excessive electromagnetic interference event occurring is 30%), then according to the formula, it can be obtained that , that is, under this working condition, the probability of control circuit failure due to over - high temperature and excessive electromagnetic interference is 1.8%. For the OR - gate logic, for example, coil failure may be caused by coil aging (denoted as ) or coil overload (denoted as ). Let the probability of the coil aging event occurring be , and the probability of the coil overload event occurring be . Then, based on the OR - gate logic, the probability of coil failure caused by any one of these two reasons can be calculated by the following formula: ; Suppose in a certain operation stage, after evaluation (i.e., the probability of the coil aging event occurring is 10%), (i.e., the probability of the coil overload event occurring is 15%), then , that is, in this operation stage, the probability of coil failure due to coil aging or coil overload is 23.5%.

[0040] For example, in a fault diagnosis, it is found that the solenoid valve cannot be opened normally (i.e., the top event occurs). Through the fault tree analysis process, first, the ambient temperature is monitored (meeting the over - high temperature event ), and the electromagnetic interference intensity = 55 (meeting the excessive electromagnetic interference event ). According to the above AND - gate logic relationship and probability calculation formula, since is 0.25 under this working condition through statistics, is 0.3, and is 0.4 through historical data statistics, then , that is, the probability of the control circuit failure due to excessive temperature and electromagnetic interference exceeding the standard under this working condition is 3%. Then, the control circuit was inspected, and it was found that a certain chip in the circuit failed (that is, it was determined that the control circuit failed occurred). According to the logical relationship of the fault tree, it can be determined that this solenoid valve failure was caused by the control circuit failure triggered by environmental factors. In order to continuously improve the accuracy of fault diagnosis, new fault cases are continuously collected. In the subsequent fault data statistics, it is found that in a high humidity ( > 80%) environment, the probability of the coil being damp and short-circuited (denoted as ) increases by 20%. Assume that the original probability of the coil being damp and short-circuited event in a general environment is , and in a high humidity environment, its occurrence probability becomes . In the fault tree, there is an association between humidity and coil failure. Let the correlation coefficient between the humidity factor (that is, the event of excessive humidity ) and the event of the coil being damp and short-circuited be . Through the analysis of 200 newly collected fault cases in a high humidity environment, it is found that 50 cases have the situation of the coil being damp and short-circuited, then . Based on this, in a high humidity environment, the probability of the coil failure caused by excessive humidity can be calculated by the following formula: ; That is, in a high humidity environment, the probability of the coil failure caused by excessive humidity is 1.5%. By continuously updating the occurrence probabilities and logical relationships of each event in the fault tree in this way, the fault diagnosis model based on the fault tree can more accurately locate the cause of the solenoid valve failure in a complex environment.

[0041] Embodiments disclosed by the present invention. The processes described above with reference to the flowcharts can be implemented as computer software programs. Embodiments disclosed by the present invention include a computer program product, which includes a computer program carried on a computer-readable medium. The computer program contains program codes for executing the methods shown in the flowcharts. In such an embodiment, the computer program can be downloaded and installed from a network through a communication part, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), the above functions defined in the methods of the present application are executed. It should be noted that the computer-readable medium in the present application can be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of a computer-readable storage medium can include, but are not limited to: an electrical connection having one or more wire segments, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present application, a computer-readable storage medium can be any tangible medium that contains or stores a program, and the program can be used by or combined with an instruction execution system, apparatus, or device. In the present application, a computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries computer-readable program codes. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, and the computer-readable medium can send, propagate, or transmit a program for use by or combined with an instruction execution system, apparatus, or device. The program codes contained on the computer-readable medium can be transmitted by any suitable medium, including but not limited to: wireless, wire segments, optical cables, RF, etc., or any suitable combination of the above.

[0042] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in the flowchart or block diagram may represent a module, a segment of a program, or a portion of code that contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions noted in the blocks may occur in a different order than that noted in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, or they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, may be implemented by a dedicated hardware-based system that performs the specified functions or operations, or may be implemented by a combination of dedicated hardware and computer instructions.

[0043] Those skilled in the art should understand that the embodiments of the present invention described above and shown in the accompanying drawings are only examples and do not limit the present invention. The objectives of the present invention have been fully and effectively achieved. The functions and structural principles of the present invention have been demonstrated and illustrated in the embodiments. Without departing from the said principles, the embodiments of the present invention may have any variations or modifications.

Claims

1. A method for detecting abnormalities in the solenoid valve of a jacquard machine, characterized in that, The method includes the following: Adopt multi-channel parallel detection to collect and record the control signal, current signal, and magnetic force signal of the solenoid valve; Based on deep learning technology, construct a model that integrates a convolutional neural network and a long short-term memory network, incorporate environmental parameters to perform feature extraction and time series analysis on the solenoid valve control signal, and measure the response time; Based on the generative adversarial network, process the extended input obtained by concatenating the input noise and the parametric linear transformation of the environmental parameters. The generator and the discriminator are trained adversarially. According to the current signal data collected by the high-precision current sensor and combined with the environmental parameters, measure the current stability; Adopt a quantum-enhanced sensor. Consider various environmental parameters through a quantum algorithm to construct a quantum mechanics model. Calibrate the measured original magnetic force signal and compare it with the standard value to measure the magnetic force intensity of the solenoid valve; Construct a fault diagnosis model based on a fault tree. Determine the logical relationship between events based on historical fault data, analyze the event probability and logic, and locate the cause of the solenoid valve fault.

2. The abnormal detection method of the solenoid valve of a jacquard machine according to claim 1, wherein, The construction of the model that integrates a convolutional neural network and a long short-term memory network includes the following steps: ； Wherein: is the input signal, is the dimension of the original input signal, is the temperature, is the humidity, is the electromagnetic interference intensity, is the dust concentration, is the chemical agent concentration, is the extended input, the output of the th convolutional layer is calculated by the following formula: ； Wherein: is the weight of the th convolutional kernel of the th convolutional layer, is the th channel of the extended input is the bias of the th convolutional layer, represents a convolutional operation; Incorporate environmental parameters into the LSTM model, and the environmental parameters are incorporated through the following transformation: ； Wherein: and are the mean and standard deviation of the temperature in the training set, and are the mean and standard deviation of the humidity in the training set, and are the mean and standard deviation of the electromagnetic interference intensity in the training set, and are the mean and standard deviation of the dust concentration in the training set, and are the mean and standard deviation of the chemical agent concentration in the training set, is the input signal to the LSTM model, is the extended input with environmental factors added.

3. The abnormal detection method for the solenoid valve of a jacquard machine according to claim 2, wherein The training process of the model that integrates a convolutional neural network and a long short-term memory network is as follows: Preset the learning rate, exponential decay rates of the first moment and the second moment; in each round of training, calculate the gradient of the model parameters according to the loss function; the Adam optimizer dynamically adjusts the model parameters according to these gradients; regularly evaluate the performance of the model on the validation set during training. If the value of the loss function does not decrease after several consecutive iterations, appropriately reduce the learning rate to prevent the model from falling into a local optimal solution; continuously loop through the training until the model reaches satisfactory accuracy and error metrics on the validation set.

4. The solenoid valve abnormality detection method for a jacquard machine according to claim 1, characterized in that, The detection process of the current stability is as follows: The input noise vector and the linearly transformed environmental parameters are concatenated into an extended input; calculate the generator output through the weight matrices and bias vectors of different layers, combined with the activation function. The ReLU activation function is used in the hidden layer, and the Tanh activation function is used in the output layer; use the current data collected by the high-precision current sensor combined with the environmental parameters as the input, optimize the performance of the generator and the discriminator through adversarial training, and use the cross-entropy loss function to measure the training effect.

5. The abnormal detection method for the solenoid valve of a jacquard machine according to claim 1, characterized in that, The calculation formula for the magnetic force intensity of the solenoid valve is as follows: ； Wherein: is the original magnetic strength, is the calibrated magnetic strength, is the temperature, is the humidity, is the electromagnetic interference strength, is the dust concentration, is the chemical agent concentration.

6. The abnormal detection method of the solenoid valve of a jacquard machine according to claim 1, characterized in that, The specific content of the fault diagnosis model is as follows: Take the solenoid valve fault as the top event, and set the coil fault, spool fault, control circuit fault, and environmental factors as intermediate and low-level events; through the analysis of historical fault data, clarify the logical relationship between each event; combined with specific fault phenomena, infer the cause of the fault based on the determined logical relationship; continuously collect new fault cases to update the event occurrence probability and logical relationship.

7. The abnormal detection method for the solenoid valve of a jacquard machine according to claim 1, characterized in that, Each channel of the multi-channel parallel detection is mutually shielded, and a filtering algorithm is added to the signal processing current. Use a high-speed voltage sensor to collect the control signal; use a high-precision current sensor to collect the current signal; Use a quantum-enhanced magnetic sensor to collect the magnetic force signal.

8. An abnormal detection system for the electromagnetic valve of a jacquard machine, characterized in that, The system is used to execute a method for detecting abnormal solenoid valves of a jacquard machine according to any one of claims 1-7.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement a method for detecting abnormal solenoid valves of a jacquard machine according to any one of the above claims 1-7.

Images