An analysis and judgment method, device and medium for monitoring filter failure fault points

Abstract

The embodiment of the application discloses a kind of analysis and judgment method, equipment and medium for monitoring filter failure fault point, belong to filter detection technical field, solve when filter carries out fault investigation, multiple filter cartridges need to be checked one by one, so as to time-consuming and labor-consuming problem.For the gas output end data sample corresponding respectively in the case where different filter cartridges of the multiple filter cartridge filter to be measured appear fault;Characteristic data extraction is carried out on the gas output end data sample, and the first feature data matrix set corresponding respectively when different filter cartridges fail is constructed based on the extracted characteristic data;The distribution correlation of the first feature data matrix set corresponding respectively when different filter cartridges fail is obtained by neural network graph;Data enhancement processing is carried out on the first feature data matrix set, and the second feature data matrix set corresponding respectively when different filter cartridges fail is obtained;Fault point detection model is constructed based on the second feature data matrix set, to carry out fault point detection for multiple filter cartridge filter.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of filter detection, and particularly relates to an analysis and judgment method, device and medium for monitoring failure fault points of a filter. BACKGROUND

[0002] In today's industrial production and environmental purification fields, multi-cylinder filters play a crucial role as a high-efficiency gas purification device. They can effectively remove particulate matter, harmful gases or other pollutants in the air through multiple built-in filter cylinders, ensuring that the discharged gas meets environmental protection standards or production process requirements. However, as the use time increases, the filter cylinders may gradually fail due to blockage, damage or aging, thereby affecting the performance and efficiency of the entire filtration system. Therefore, timely and accurate detection of the failure state of the multi-cylinder filter is of great significance for maintaining production safety, environmental protection and cost control.

[0003] Existing multi-cylinder failure detection methods mostly rely on installing sensors at the gas output end to monitor the concentration changes of the output gas in real time. However, when the existing system detects filter failure, maintenance personnel often take a one-by-one troubleshooting approach to check the condition of each filter cylinder. This process not only consumes time and effort, reducing the timeliness of fault response, but also increases the risk of filter damage due to frequent disassembly and assembly operations, further increasing maintenance costs. SUMMARY

[0004] The embodiments of the present application provide an analysis and judgment method, device and medium for monitoring failure fault points of a filter, which are used to solve the following technical problems: when troubleshooting the filter, maintenance personnel often troubleshoot multiple filter cylinders one by one. This process not only consumes time and effort, reducing the timeliness of fault response, but also increases the risk of filter damage due to frequent disassembly and assembly operations, further increasing maintenance costs.

[0005] The embodiments of the present application adopt the following technical solutions:

[0006] The embodiment of the application provides a kind of monitoring filter failure fault point analysis and judgment method, comprising, obtain the corresponding gas output end data sample of different filter cartridge of the multiple filter cartridge filter to be tested in the case where fault occurs;Wherein, multiple gas detection devices are uniformly arranged in gas output end;Multiple gas detection devices correspond to detecting different filter cartridge areas respectively;Characteristic data extraction is carried out to gas output end data sample, and first feature data matrix set corresponding to different filter cartridge faults is constructed based on the extracted feature data;First feature data matrix set is input into neural network graph, and first feature data matrix set distribution association relationship corresponding to different filter cartridge faults is obtained by neural network graph;Based on first feature data matrix set distribution association relationship, data enhancement processing is carried out to first feature data matrix set, and second feature data matrix set corresponding to different filter cartridge faults is obtained;Fault point detection model is constructed based on second feature data matrix set, to carry out fault point detection to multiple filter cartridge filter based on fault point detection model and current obtained gas output end data.

[0007] The embodiment of the application can accurately analyze and compare the data characteristics of different filter cartridge faults by constructing the feature data matrix set, which provides convenience for subsequent data processing and model training. The potential association relationship between different filter cartridge faults is determined by processing the feature data matrix set through the neural network, which provides a more accurate basis for subsequent fault point detection. The diversity and richness of data are increased by data enhancement processing, which further improves the generalization ability and robustness of the model, so that it can more accurately identify fault points when facing new data. By constructing the model based on the second feature data matrix set, the extracted features and association relationships are fully utilized to accurately detect the fault points of the multiple filter cartridge filter. This not only improves the efficiency and accuracy of fault detection, but also reduces maintenance cost and time cost.

[0008] In one implementation of the application, the feature data of the gas output end data sample is extracted, and the first feature data matrix set corresponding to different filter cartridge faults is constructed based on the extracted feature data, specifically including: time domain feature extraction is carried out to the gas output end data sample;And by Hilbert transform, gas concentration feature extraction is carried out to the gas output end data sample;Based on feature type, time domain feature and gas concentration feature are divided;Wherein, time domain feature at least includes one of peak time, root mean square value and waveform factor;Gas concentration feature at least includes instantaneous amplitude extraction value and gas concentration frequency extraction value;When the current filter cartridge to be tested fails, different types of features corresponding to multiple gas detection devices are determined, and multiple feature data matrices are constructed based on different types of features, to obtain the first feature data matrix set corresponding to the current fault filter cartridge.

[0009] In an implementation form of the present application, the first feature data matrix set is input into the neural network graph, and the distribution correlation of the first feature data matrix set corresponding to different filter cartridge faults is obtained through the neural network graph. Specifically, the first feature data matrix set corresponding to the first time period is input into the graph neural network. In the graph neural network, the gas concentration difference between each preset filter cartridge key point corresponding to the first time period is determined. The preset filter cartridge key points are uniformly arranged in the graph neural network, and the positions of the preset filter cartridge key points are related to the actual positions of the filter cartridges. The first feature data matrix set corresponding to the next time period is input into the graph neural network. In the graph neural network, the gas concentration difference between each preset filter cartridge key point corresponding to the next time period is determined. Until the first feature data matrix set is input into the graph neural network according to the time sequence of the gas output end data sample extraction, the gas concentration difference corresponding to multiple time periods is obtained. The key point relationship score is determined based on the gas concentration difference, and the distribution correlation of the first feature data matrix set corresponding to different filter cartridge faults is determined based on the change of the gas concentration difference and the key point relationship score corresponding to multiple time periods.

[0010] In an implementation form of the present application, the distribution correlation of the first feature data matrix set corresponding to different filter cartridge faults is determined based on the change of the gas concentration difference and the key point relationship score corresponding to multiple time periods. Specifically, the node distance between each preset filter cartridge key point is determined, and the first node score is obtained based on the node distance. The second node score is determined according to the gas concentration difference between each preset filter cartridge key point. The key point relationship score is obtained based on the first node score and the second node score. The gas concentration change curve between each preset filter cartridge key point is constructed based on the gas concentration difference corresponding to multiple time periods. The key point relationship score change curve between each preset filter cartridge key point is constructed based on the key point relationship score corresponding to multiple time periods. The distribution correlation of the first feature data matrix set corresponding to different filter cartridge faults is determined based on the gas concentration change curve and the relationship score change curve.

[0011] In an implementation manner of the present application, the first feature data matrix set is subjected to data enhancement processing based on the distribution association relationship of the first feature data matrix set, specifically including: performing normalization processing on the first feature data matrix set to adjust the scale of the feature data; performing feature polynomial combination processing on a plurality of feature data matrices corresponding to the same fault filter cartridge to generate new features; inputting the distribution association relationship of the first feature data matrix set and the new features into the preset GANs model to generate first new data samples; inputting the distribution association relationship of the first feature data matrix set and the new features into the preset trained VAEs model to generate second new data samples; and performing data enhancement processing on the first feature data matrix set based on the new features, the first new data samples and the second new data samples.

[0012] In an implementation manner of the present application, the fault point detection model is constructed based on the second feature data matrix set, specifically including: training the preset neural network model based on the second feature data matrix set; determining the derived vector group corresponding to each second feature data matrix set through a global self-attention mechanism, performing multiplication and summation processing on the derived vector group to obtain the global most relevant information corresponding to each second feature data matrix set, and adjusting the preset neural network model based on the global most relevant information; updating the weights of the adjusted preset neural network model through a back propagation algorithm; and obtaining the fault point detection model when the training result meets the requirements.

[0013] In an implementation manner of the present application, before the fault point detection model and the currently acquired gas output end data are used to detect the fault point of the multi-filter cartridge filter, the method further includes: inputting the currently acquired gas output end data into the first time sequence detection model to obtain the time feature sequence of the to-be-detected multi-filter cartridge filter in the current time period; performing frequency domain feature extraction on the currently acquired gas output end data, inputting the frequency domain features into the second time sequence prediction model to obtain the frequency domain feature sequence corresponding to the to-be-detected multi-filter cartridge filter in the current time period; acquiring the gas component data corresponding to the gas inlet end of the to-be-detected multi-filter cartridge filter in the current time period, determining a warning threshold based on the gas component data; acquiring the total number of the time features in the time feature sequence and the frequency domain features in the frequency domain feature sequence, and generating warning information when the ratio between the number of abnormal features and the total number is greater than the warning threshold.

[0014] In an implementation manner of the present application, the fault point of the multi-filter cartridge filter is detected based on the fault point detection model and the currently acquired gas output end data, specifically including: inputting the currently acquired gas output end data into the fault point detection model to output a plurality of reference fault filter cartridge information based on the fault point detection model; and performing fault alarm based on the priority of the plurality of reference fault filter cartridge information.

[0015] The embodiment of the application provides an analysis and judgment device for monitoring filter failure fault points, comprising: at least one processor; and a memory in communication connection with the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to: acquire gas output end data samples corresponding to different filter cartridges of a to-be-tested multi-filter cartridge filter respectively in the case that the different filter cartridges fail; wherein a plurality of gas detection devices are uniformly arranged at the gas output end; the plurality of gas detection devices correspond to different filter cartridge areas respectively; feature data of the gas output end data samples is extracted, a first feature data matrix set corresponding to different filter cartridge failures respectively is constructed based on the extracted feature data; the first feature data matrix set is input into a neural network graph, and a distribution correlation of the first feature data matrix set corresponding to different filter cartridge failures respectively is obtained through the neural network graph; the first feature data matrix set is subjected to data enhancement processing based on the distribution correlation of the first feature data matrix set, and a second feature data matrix set corresponding to different filter cartridge failures respectively is obtained; and a fault point detection model is constructed based on the second feature data matrix set, so that the multi-filter cartridge filter is subjected to fault point detection based on the fault point detection model and currently acquired gas output end data.

[0016] The embodiment of the application provides a nonvolatile computer storage medium, which stores computer executable instructions, and the computer executable instructions are set to: acquire gas output end data samples corresponding to different filter cartridges of a to-be-tested multi-filter cartridge filter respectively in the case that the different filter cartridges fail; wherein a plurality of gas detection devices are uniformly arranged at the gas output end; the plurality of gas detection devices correspond to different filter cartridge areas respectively; feature data of the gas output end data samples is extracted, a first feature data matrix set corresponding to different filter cartridge failures respectively is constructed based on the extracted feature data; the first feature data matrix set is input into a neural network graph, and a distribution correlation of the first feature data matrix set corresponding to different filter cartridge failures respectively is obtained through the neural network graph; the first feature data matrix set is subjected to data enhancement processing based on the distribution correlation of the first feature data matrix set, and a second feature data matrix set corresponding to different filter cartridge failures respectively is obtained; and a fault point detection model is constructed based on the second feature data matrix set, so that the multi-filter cartridge filter is subjected to fault point detection based on the fault point detection model and currently acquired gas output end data.

[0017] The at least one technical solution adopted by the embodiments of the present application can achieve the following beneficial effects: The embodiments of the present application can more accurately locate the faulty filter cartridge by uniformly arranging multiple detection devices at the gas output end and independently monitoring different filter cartridge regions, thereby reducing the possibility of false positives and false negatives. By constructing a feature data matrix set, the data features of different filter cartridges when a fault occurs can be accurately analyzed and compared, thereby facilitating subsequent data processing and model training. By processing the feature data matrix set through a neural network, the potential correlation between different filter cartridge faults can be determined, thereby providing a more accurate basis for subsequent fault point detection. By increasing the diversity and richness of data through data augmentation processing, the generalization ability and robustness of the model are further improved, so that the model can more accurately identify fault points when facing new data. By constructing a model based on the second feature data matrix set, the extracted features and correlation are fully utilized to accurately detect the fault points of the multi-filter cartridge filter. This not only improves the efficiency and accuracy of fault detection, but also reduces maintenance costs and time costs. BRIEF DESCRIPTION OF DRAWINGS

[0018] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can be obtained by those skilled in the art without creative labor. In the drawings:

[0019] Figure 1 A flow chart of an analysis and judgment method for monitoring filter failure fault points is provided for the embodiments of the present application.

[0020] Figure 2 A structural schematic diagram of an analysis and judgment device for monitoring filter failure fault points is provided for the embodiments of the present application.

[0021] Reference signs:

[0022] 200: an analysis and judgment device for monitoring filter failure fault points, 201: a processor, 202: a memory. DETAILED DESCRIPTION

[0023] The embodiments of the present application provide an analysis and judgment method, device and medium for monitoring filter failure fault points.

[0024] In order for those skilled in the art to better understand the technical solutions in the present application, the technical solutions in the embodiments of the present application will be clearly and completely described in the following with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments of the present application, all other embodiments obtained by those skilled in the art without creative labor should fall within the scope of protection of the present application.

[0025] The technical solutions proposed by the embodiments of the present application will be described in detail below with reference to the drawings.

[0026] Figure 1 The flow chart of the analysis and judgment method for monitoring the failure fault point of the filter provided by the embodiments of the present application is shown as follows. Figure 1 The analysis and judgment method for monitoring the failure fault point of the filter includes the following steps:

[0027] S101, acquiring the gas output end data samples corresponding to different filter cartridges respectively in the case of failure of different filter cartridges of the multi-filter cartridge filter.

[0028] In an embodiment of the present application, multiple gas detection devices are uniformly arranged at the gas output end of the multi-filter cartridge filter; the multiple gas detection devices correspond to detection of different filter cartridge regions respectively.

[0029] In an embodiment of the present application, multiple gas detection devices need to be uniformly arranged at the gas output end of the multi-filter cartridge filter. The positions of these devices should ensure that they can comprehensively cover the exhaust regions of all filter cartridges, so as to accurately detect the working state of each filter cartridge. Each gas detection device should correspond to detection of different filter cartridge regions, usually by dividing the output end into multiple independent detection regions, each region corresponding to one or more filter cartridges.

[0030] Further, in order to acquire data samples when different filter cartridges fail, various fault conditions need to be simulated. For example, a filter cartridge can be artificially blocked, its filter medium can be damaged, or its working pressure can be adjusted, etc. to simulate real fault scenarios. In the process of simulating faults, the gas detection devices are used to record the data of the gas output end in real time. These data should include key parameters such as gas concentration, flow rate, temperature, etc. to comprehensively reflect the working state of the filter cartridge. Each data sample is labeled to indicate the corresponding filter cartridge and fault type. For example, numbers or letter codes can be used to represent different filter cartridges and fault types, so as to be identified in subsequent data analysis and model training.

[0031] S102, feature data extraction is performed on the gas output end data samples, and a first feature data matrix set corresponding to different filter cartridge failures respectively is constructed based on the extracted feature data.

[0032] In one embodiment of the present application, time domain feature extraction is performed on the gas output end data sample; and gas concentration feature extraction is performed on the gas output end data sample through Hilbert transform. The time domain features and the gas concentration features are divided based on feature types; wherein the time domain features at least include one of peak time, root mean square value and waveform factor; and the gas concentration features at least include instantaneous amplitude extraction value and gas concentration frequency extraction value. When the current filter cartridge to be tested is faulty, different types of features corresponding to the plurality of gas detection devices are determined, and a plurality of feature data matrices are constructed based on the different types of features to obtain a first feature data matrix set corresponding to the current filter cartridge to be tested.

[0033] Specifically, the time domain features extracted in the embodiments of the present application can include peak time, root mean square value and waveform factor. The peak time is the time point at which the signal reaches the maximum value, and in gas concentration monitoring, the peak time represents the peak time of pollutant emission; the root mean square value is the square root of the average value of the square sum of the signal amplitude, which can reflect the energy size of the signal, and for gas concentration data, it represents the average concentration level of the pollutant; and the waveform factor can reflect the waveform characteristics of the signal, such as sharpness or flatness.

[0034] Further, the instantaneous amplitude through Hilbert transform reflects the instantaneous change of gas concentration, and through frequency domain analysis of the instantaneous amplitude, the frequency components or feature values obtained reflect the frequency characteristics of the change of gas concentration with time. For each gas detection device, a set of time domain features and gas concentration features can be extracted, and these features can be organized according to feature types to construct a first feature data matrix set. For example, in the case of the current filter cartridge to be tested being faulty, the instantaneous amplitude extraction values corresponding to each detection device are organized to obtain the corresponding feature data matrix, and the instantaneous amplitude extraction values change with time, so that a plurality of time-continuous feature data matrices can be obtained within the current time period to construct the first feature data matrix set corresponding to the current filter cartridge to be tested.

[0035] S103, input the first feature data matrix set into the neural network graph to obtain the distribution association relationship of the first feature data matrix set corresponding to different filter cartridge faults through the neural network graph.

[0036] In an embodiment of the present application, the first feature data matrix set corresponding to the first time period is input into the graph neural network. In the graph neural network, the gas concentration difference values between each preset filter key point corresponding to the first time period are determined; wherein the preset filter key points are uniformly arranged in the graph neural network, and the positions of the preset filter key points are related to the actual positions of each filter. The first feature data matrix set corresponding to the next time period is input into the graph neural network. In the graph neural network, the gas concentration difference values between each preset filter key point corresponding to the next time period are determined. Until the first feature data matrix set is input into the graph neural network in the order of the time sequence of the gas output end data sample extraction, the gas concentration difference values corresponding to multiple time periods are obtained respectively. Based on the gas concentration difference values, the key point relationship scores are determined, and based on the changes of the gas concentration difference values and the key point relationship scores corresponding to multiple time periods respectively, the distribution association relationship of the first feature data matrix set corresponding to different filter failures respectively is determined.

[0037] Specifically, in the graph neural network, the preset filter key points are related to the positions of each filter, the edges between nodes represent the spatial relationship or connection between filters, and the feature vector of the node is composed of the feature values in the first feature data matrix set. The first feature data matrix set of each time period is input into the graph neural network as input, and each feature data matrix contains the time domain features and gas concentration features of all filters in the time period.

[0038] Further, the features related to the gas concentration are extracted from the node feature vector of the graph neural network, such as the instantaneous amplitude extraction value, and for each time period, the gas concentration difference values between each preset filter key point (i.e. node) are calculated, which reflect the relative changes of the gas concentration between the filters. According to the time sequence of the gas output end data sample extraction, the first feature data matrix set of each time period is input into the graph neural network in turn, and for each input time period, the graph neural network outputs the gas concentration difference values between the filter key points in the time period.

[0039] Further, based on the gas concentration difference values, the relationship scores between the filter key points in each time period can be calculated, which reflect the degree of mutual influence or the strength of association between the filters. Based on the changes of the gas concentration difference values and the key point relationship scores in multiple time periods, the distribution association relationship of the first feature data matrix set corresponding to different filter failures respectively is determined.

[0040] In an embodiment of the present application, the node distance between each preset filter key point is determined, and a first node score is obtained based on the node distance. A second node score is determined based on the gas concentration difference between each preset filter key point. A key point relationship score is obtained based on the first node score and the second node score. A gas concentration change curve between each preset filter key point is constructed based on the gas concentration difference corresponding to each time period, and a key point relationship score change curve between each preset filter key point is constructed based on the key point relationship score corresponding to each time period. Based on the gas concentration change curve and the relationship score change curve, a distribution association relationship of the first feature data matrix set corresponding to different filter faults is determined.

[0041] Specifically, in the graph neural network, each preset filter key point (node) represents a filter or a key position of the filter. A score, i.e., a first node score, is assigned to each node based on the node distance. For each time period, the gas concentration difference between each preset filter key point is calculated, and another score, i.e., a second node score, is assigned to each node based on the gas concentration difference. In combination with the first node score and the second node score, a key point relationship score between each preset filter key point in each time period can be obtained.

[0042] Further, a gas concentration change curve between each preset filter key point is constructed based on the gas concentration difference corresponding to each time period, which shows the trend of the change of the gas concentration over time. Similarly, a key point relationship score change curve between each preset filter key point is constructed based on the key point relationship score corresponding to each time period, which shows the change of the key point relationship score over time. In combination with the gas concentration change curve and the key point relationship score change curve, the association relationship between different filters and their changes in different time periods can be more comprehensively understood.

[0043] For example, assuming that the distance between filter 1 and filter 2 is closer, and the distance between filter 3 and them is farther, therefore, the first node score of filter 1 and filter 2 is higher, and the first node score of filter 3 is lower. In time period 1, the gas concentration difference between filter 1 and filter 2 is smaller, and the difference between filter 3 and them is larger, therefore, the second node score of filter 1 and filter 2 is higher, and the second node score of filter 3 is lower. In combination with the first node score and the second node score, the key point relationship score between each filter key point in each time period is calculated. The gas concentration change curve and the key point relationship score change curve are constructed, and through the curves, it can be obtained that the gas concentration of filter 3 suddenly rises in time period 3, and its key point relationship score with other filters also significantly decreases. Based on the analysis of these curves, it is determined that filter 3 has a fault in time period 3, and this fault may affect the performance of other filters.

[0044] S104, based on the distribution association relationship of the first feature data matrix set, performing data enhancement processing on the first feature data matrix set to obtain a second feature data matrix set corresponding to different filter cartridge faults respectively.

[0045] In an embodiment of the present application, the first feature data matrix set is normalized to adjust the scale of the feature data. The feature polynomial combination processing is performed on the plurality of feature data matrices corresponding to the same fault filter cartridge to generate a new feature. The first feature data matrix set distribution association relationship and the new feature are input into the preset GANs model to generate a first new data sample. The first feature data matrix set distribution association relationship and the new feature are input into the preset trained VAEs model to generate a second new data sample. Based on the new feature, the first new data sample and the second new data sample, the data enhancement processing is performed on the first feature data matrix set.

[0046] Specifically, the first feature data matrix set is normalized, and each feature value is adjusted to a specific range. For the plurality of feature data matrices corresponding to the same fault filter cartridge, two features (such as gas concentration and temperature) can be selected for polynomial combination processing. For example, a new feature "gas concentration * temperature" is generated, which combines different features and can better reflect the working state and fault condition of the filter cartridge.

[0047] Further, the normalized first feature data matrix set distribution association relationship and the new feature are input into the preset GANs model. The generator of the GANs generates a first new data sample according to the input information. The sample is similar to the original data in feature distribution, but has new details and changes. Similarly, these information is input into the preset VAEs model to generate a second new data sample through the decoding process. Based on the new feature, the first new data sample and the second new data sample, the data enhancement processing can be performed on the original first feature data matrix set. For example, the new data samples can be mixed with the original data, or they can be combined in some way (such as data fusion, feature splicing, etc.) to form a more rich data set. Thus, the enhanced data set can be used to train a more accurate model to identify the fault mode of the filter cartridge.

[0048] Through the data preprocessing and enhancement steps, the embodiments of the present application can improve the quality and richness of the data, thereby improving the performance and generalization ability of the model.

[0049] S105, based on the second feature data matrix set, constructing a fault point detection model, and based on the fault point detection model and the current acquired gas output end data, performing fault point detection on the multi-filter filter.

[0050] In an embodiment of the present application, the preset neural network model is trained based on the second feature data matrix set. Through the global self-attention mechanism, the derived vector group corresponding to each second feature data matrix set is determined, and the derived vector group is processed by multiplication and summation to obtain the global most relevant information corresponding to each second feature data matrix set, so as to adjust the preset neural network model based on the global most relevant information. The preset neural network model after adjustment is updated in weight through the back propagation algorithm. In the case that the training result meets the requirement, the fault point detection model is obtained.

[0051] Specifically, the preset neural network model is trained by the second feature data matrix set obtained after data enhancement. The preset neural network model in the embodiment of the present application can be a convolutional neural network.

[0052] Further, the global self-attention mechanism is introduced in the embodiment of the present application to determine the most relevant information corresponding to each second feature data matrix set. The global self-attention mechanism is a mechanism that can capture the relationship between all elements in the data set. It can allocate weights according to the correlation between elements, so as to highlight the most important information. For each second feature data matrix set, it is converted into a derived vector group. Then, the global self-attention mechanism is used to calculate the correlation between these vectors and obtain a weighted sum, i.e. the global most relevant information. Assuming that the second feature data matrix set contains gas concentration data at multiple time points, these data are converted into a derived vector group, and the global self-attention mechanism is used to calculate the correlation between these vectors. Through calculation, a weighted sum is obtained, which represents the global most relevant information of gas concentration at all time points.

[0053] Further, the global most relevant information is used to adjust the parameters of the preset neural network model. This can be achieved by taking the global most relevant information as an additional input or a component of the loss function. By adjusting the model parameters, the model can pay more attention to the global most relevant information, thereby improving the performance of the model. For example, when training a CNN model, the global most relevant information can be added as an additional input feature to the input layer of the model, or it can be used as a component of the loss function to optimize the model. In this way, the model will pay more attention to the global most relevant information during training, thereby improving the detection ability of the fault point.

[0054] Further, the weights of the pre-trained neural network model are updated using a backpropagation algorithm. The backpropagation algorithm is an optimization algorithm used to train neural networks, which updates the weights by calculating the gradient of the loss function with respect to the model parameters. When training a CNN model, the backpropagation algorithm is used to calculate the gradient of the loss function with respect to the model weights, and the weights are updated based on these gradients. Through multiple iterations of training, the loss value of the model can be gradually reduced, thereby improving the performance of the model. If the performance of the model reaches a predetermined standard (e.g., the accuracy, recall rate, and other indicators reach a certain threshold), it can be considered that the training process has been completed, and a fault point detection model is obtained.

[0055] In an embodiment of the present application, the current acquired gas output end data is input into a first time sequence detection model to obtain a time feature sequence of the to-be-tested multi-filter cartridge filter in the current time period, and frequency domain feature extraction is performed on the current acquired gas output end data, and the frequency domain features are input into a second time sequence prediction model to obtain a frequency domain feature sequence corresponding to the current time period of the to-be-tested multi-filter cartridge filter. The gas component data corresponding to the gas inlet end of the to-be-tested multi-filter cartridge filter in the current time period is acquired, and a warning threshold is determined based on the gas component data. The total number of time features in the time feature sequence and frequency domain features in the frequency domain feature sequence is acquired, and in a case where a ratio between the number of abnormal features and the total number is greater than the warning threshold, a warning information is generated.

[0056] Specifically, the first time sequence detection model in the embodiment of the present application can be an LSTM model, and the current acquired gas output end data is input into the LSTM model, and the model outputs a time feature sequence, wherein each feature represents the state change of the filter at a certain time point, such as flow fluctuation, pressure change rate, etc.

[0057] Further, the gas output end data is subjected to FFT transformation to extract frequency components such as fundamental frequency, harmonic, etc. Then, these frequency domain features are input into another time sequence prediction model such as an ARIMA model or a prediction model based on deep learning, and the model outputs a frequency domain feature sequence, which contains the change of different frequency components over time.

[0058] Further, the gas component data corresponding to the gas inlet end of the to-be-tested multi-filter cartridge filter in the current time period is acquired. Based on these gas component data, a warning threshold is determined for subsequent judgment of whether the number of abnormal features reaches a warning condition. It is assumed that the monitored gas component includes oxygen content. According to historical data and experience, it is set that when the oxygen content is lower than a certain threshold (such as lower than 20%), the filter may face the risk of blockage or failure. Therefore, this threshold of oxygen content is taken as the warning threshold.

[0059] Further, the total number of time features in the time feature sequence and frequency domain features in the frequency domain feature sequence is obtained. Then, the number of abnormal features (i.e., those exceeding the normal range or expected number of features) is calculated. If the ratio between the number of abnormal features and the total number is greater than the early warning threshold, early warning information is generated, indicating that the filter may have a problem or needs maintenance.

[0060] For example, 100 time features and 100 frequency domain features are obtained, a total of 200 features. Through analysis, it is found that 30 of them are outside the normal range (such as abnormally large flow fluctuations, abnormal frequency components, etc.). Therefore, the number of abnormal features is 30. If the set early warning threshold is 10%, then since 30 / 200 = 15% is greater than the early warning threshold, it is considered that the filter may be abnormal, and early warning information is generated.

[0061] In an embodiment of the present application, the current obtained gas output end data is input into the fault point detection model to output a plurality of reference fault filter cartridge information based on the fault point detection model. Based on the priority of the plurality of reference fault filter cartridge information, a fault alarm is performed.

[0062] Specifically, after obtaining the current gas output end data, it is input into the fault point detection model, and the fault point detection model will output a plurality of reference fault filter cartridge information based on the input gas output end data. The importance of each fault point is determined based on the probability, degree, impact on the system, or emergency degree of the fault, and a fault alarm is performed according to the priority.

[0063] Figure 2 A structural schematic diagram of an analysis and judgment device for monitoring filter failure fault points is provided for an embodiment of the present application. As shown in Figure 2As shown, the analysis and judgment device 200 for monitoring filter failure fault points comprises: at least one processor 201; and a memory 202 in communication connection with the at least one processor 201; wherein the memory 202 stores instructions executable by the at least one processor 201, and the instructions are executed by the at least one processor 201 to enable the at least one processor 201 to: acquire gas output end data samples corresponding to different filter cartridges in the case of failure of the different filter cartridges of a multi-filter cartridge filter to be tested; wherein a plurality of gas detection devices are uniformly arranged at intervals on the gas output end; the plurality of gas detection devices correspond to detection of different filter cartridge regions; feature data is extracted from the gas output end data samples, and a first feature data matrix set corresponding to different filter cartridge failures is constructed based on the extracted feature data; the first feature data matrix set is input into a neural network graph, and a distribution correlation relationship of the first feature data matrix set corresponding to different filter cartridge failures is obtained through the neural network graph; the first feature data matrix set is subjected to data enhancement processing based on the distribution correlation relationship of the first feature data matrix set, and a second feature data matrix set corresponding to different filter cartridge failures is obtained; and a fault point detection model is constructed based on the second feature data matrix set, so as to detect fault points of the multi-filter cartridge filter based on the fault point detection model and the currently acquired gas output end data.

[0064] The non-volatile computer storage medium provided by the embodiments of the present application stores computer executable instructions, and the computer executable instructions are configured to: acquire gas output end data samples corresponding to different filter cartridges in the case of failure of the different filter cartridges of a multi-filter cartridge filter to be tested; wherein a plurality of gas detection devices are uniformly arranged at intervals on the gas output end; the plurality of gas detection devices correspond to detection of different filter cartridge regions; feature data is extracted from the gas output end data samples, and a first feature data matrix set corresponding to different filter cartridge failures is constructed based on the extracted feature data; the first feature data matrix set is input into a neural network graph, and a distribution correlation relationship of the first feature data matrix set corresponding to different filter cartridge failures is obtained through the neural network graph; the first feature data matrix set is subjected to data enhancement processing based on the distribution correlation relationship of the first feature data matrix set, and a second feature data matrix set corresponding to different filter cartridge failures is obtained; and a fault point detection model is constructed based on the second feature data matrix set, so as to detect fault points of the multi-filter cartridge filter based on the fault point detection model and the currently acquired gas output end data.

[0065] Each of the embodiments in the present application is described in a progressive manner, and the same or similar parts of each of the embodiments can be referred to each other. Each of the embodiments focuses on the difference from other embodiments. In particular, for the device, equipment, and non-volatile computer storage medium embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the part of the description of the method embodiments.

[0066] The above merely provides an example of the present application, but is not intended to limit the present application. The example of the present application can be modified and changed by those skilled in the art. The modifications or replacements do not make the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the examples of the present application.

Claims

1. A method for analyzing and judging the failure points of a filter, characterized in that, The method includes: Data samples of gas output from different filter cartridges of a multi-cartridge filter under test are obtained, respectively, when different filter cartridges malfunction. Each gas output is equipped with multiple gas detection devices spaced evenly at different intervals, and each gas detection device is assigned to a different filter cartridge region. Feature data is extracted from the gas output data sample, and a first feature data matrix set corresponding to different filter cartridge failures is constructed based on the extracted feature data. The first feature data matrix set is input into the neural network graph, and the distribution correlation of the first feature data matrix set corresponding to different filter cartridge failures is obtained through the neural network graph. Based on the distribution and correlation of the first feature data matrix set, data augmentation processing is performed on the first feature data matrix set to obtain the second feature data matrix set corresponding to different filter cartridge failures. A fault detection model is constructed based on the second feature data matrix set, and fault points are detected in the multi-cartridge filter based on the fault detection model and the currently acquired gas output data. The first feature data matrix set is input into a neural network graph. The neural network graph is used to obtain the distribution correlation of the first feature data matrix set corresponding to different filter cartridge failures, specifically including: Input the first feature data matrix set corresponding to the first time period into the graph neural network; In the graph neural network, the gas concentration difference between key points of each preset filter cartridge corresponding to the first time period is determined; wherein, the key points of the preset filter cartridge are uniformly arranged in the graph neural network, and the position of the key points of the preset filter cartridge is related to the actual position of each filter cartridge; Input the first feature data matrix set corresponding to the next time period into the graph neural network; In the graph neural network, the gas concentration difference between key points of each preset filter cartridge corresponding to the next time period is determined; Until the first feature data matrix set is input into the graph neural network according to the order of extraction time of the gas output data samples, the gas concentration difference corresponding to multiple time periods is obtained. Based on the gas concentration difference, the key point relationship score is determined. Based on the changes in the gas concentration difference and the key point relationship score corresponding to the multiple time periods, the distribution correlation of the first feature data matrix set corresponding to different filter cartridge failures is determined. Based on the changes in the gas concentration differences and key point relationship scores corresponding to the multiple time periods, the distribution correlation of the first feature data matrix set corresponding to different filter cartridge failures is determined, specifically including: Determine the node distance between the key points of each of the preset filter cartridges, and obtain the first node score based on the node distance; The second node score is determined based on the gas concentration difference between the key points of each of the pre-set filter cartridges; The key point relationship score is obtained based on the first node score and the second node score; Based on the gas concentration differences corresponding to multiple time periods, gas concentration change curves were constructed between key points of each pre-set filter cartridge; and Based on the key point relationship scores corresponding to multiple time periods, construct the key point relationship score change curves between key points of each preset filter cartridge; Based on the gas concentration change curve and the relationship score change curve, the distribution correlation of the first feature data matrix set corresponding to different filter cartridge failures is determined.

2. The method for analyzing and judging the failure points of a monitoring filter according to claim 1, characterized in that, The step of extracting feature data from the gas output data sample and constructing a first feature data matrix set corresponding to different filter cartridge failures based on the extracted feature data specifically includes: Temporal feature extraction is performed on the gas output data samples; and Gas concentration features were extracted from the gas output data samples using Hilbert transform. Based on feature type, the time-domain features and the gas concentration features are divided; wherein, the time-domain features include at least one of peak time, root mean square value, and waveform factor; the gas concentration features include at least instantaneous amplitude extracted value and gas concentration frequency extracted value; When the filter cartridge under test is faulty, the different types of features corresponding to the multiple gas detection devices are determined, and multiple feature data matrices are constructed based on the different types of features to obtain the first feature data matrix set corresponding to the current faulty filter cartridge under test.

3. The method for analyzing and judging the failure points of a monitoring filter according to claim 1, characterized in that, The step of performing data augmentation processing on the first feature data matrix set based on the distribution and correlation relationships of the first feature data matrix set specifically includes: The first feature data matrix set is normalized to adjust the scale of the feature data; Multiple feature data matrices corresponding to the same faulty filter cartridge are combined using feature polynomials to generate new features. The distribution correlation of the first feature data matrix set is used to input the new features into a pre-set GANs model to generate the first new data sample; The distribution correlation of the first feature data matrix set is used to pre-train a VAEs model with the new feature input to generate a second new data sample; Based on the new features, the first new data sample, and the second new data sample, data augmentation processing is performed on the first feature data matrix set.

4. The method for analyzing and judging the failure points of a monitoring filter according to claim 1, characterized in that, The construction of the fault point detection model based on the second feature data matrix set specifically includes: The pre-set neural network model is trained based on the second feature data matrix set; By using a global self-attention mechanism, the derived vector groups corresponding to each second feature data matrix set are determined. The derived vector groups are multiplied and summed to obtain the global most relevant information corresponding to each second feature data matrix set. The preset neural network model is then adjusted based on the global most relevant information. The weights of the adjusted preset neural network model are updated using the backpropagation algorithm; If the training results meet the requirements, the fault point detection model is obtained.

5. The method for analyzing and judging the failure points of a monitoring filter according to claim 1, characterized in that, Before performing fault point detection on the multi-cartridge filter based on the fault point detection model and the currently acquired gas output data, the method further includes: The currently acquired gas output data is input into the first time series detection model to obtain the time feature sequence of the multi-filter cartridge under test within the current time period; and Frequency domain features are extracted from the currently acquired gas output data, and the frequency domain features are input into the second time series prediction model to obtain the frequency domain feature sequence of the multi-filter cartridge filter under test in the current time period. Obtain the gas composition data corresponding to the air inlet end of the multi-filter cartridge under test within the current time period, and determine the warning threshold based on the gas composition data; The total number of time features in the time feature sequence and the total number of frequency features in the frequency feature sequence are obtained. If the ratio between the number of abnormal features and the total number is greater than the warning threshold, a warning message is generated.

6. The method for analyzing and judging the failure points of a monitoring filter according to claim 1, characterized in that, The step of detecting fault points in the multi-cartridge filter based on the fault point detection model and the currently acquired gas output data specifically includes: The currently acquired gas output data is input into the fault point detection model, so as to output multiple reference fault filter cartridge information based on the fault point detection model; Fault alarms are generated based on the priority of multiple reference fault filter cartridge information.

7. An analytical and judgment device for monitoring filter failure points, characterized in that, The device includes a memory for storing computer program instructions and a processor for executing the program instructions, wherein when the computer program instructions are executed by the processor, the device is triggered to perform the method described in any one of claims 1-6.

8. A non-volatile computer storage medium storing computer-executable instructions, characterized in that, The computer-executable instructions are capable of performing the method described in any one of claims 1-6.

Images