Environmental protection facility state monitoring and diagnosing method and system based on vibration and infrared detection

Abstract

The invention discloses an environmental protection facility state monitoring and diagnosing method and system based on vibration and infrared detection, and the method comprises the following steps: collecting vibration and temperature data of environmental protection facility equipment, removing noise and environment interference through a filtering algorithm, and carrying out the standardization processing; extracting features of the data through a space-time convolutional network; performing deep fusion on the features by using an adaptive feature transformation network; performing abnormal mode detection; analyzing potential reasons of the abnormal mode by using a causal inference model, and generating a fault diagnosis report; a real-time processing strategy is generated, and the operation state of the environmental protection facility equipment is controlled in real time; and the fault diagnosis result and the instant processing strategy are uploaded to a cloud management platform, a notification is sent to related maintenance personnel, and remote monitoring and collaborative maintenance are realized. According to the method, the space-time convolutional network and the causal inference model are adopted, the faults of the environmental protection facility equipment are effectively identified, a proper emergency processing strategy is generated according to the real-time monitoring data, and the accuracy and response efficiency of equipment fault diagnosis are improved.

Description

Technical Field

[0001] The present invention relates to the field of environmental protection facility status monitoring, and in particular to an environmental protection facility status monitoring and diagnosis method and system based on vibration and infrared detection. Background Art

[0002] With the development of modern industry and the improvement of environmental protection requirements, environmental protection facilities and equipment play an increasingly important role in industrial production. These equipment are widely used in sewage treatment, waste gas treatment, noise control and other fields, aiming to reduce pollution to the environment and ensure the sustainability of the production process. However, with the increase in equipment service life and load changes, the probability of equipment failure and abnormal conditions gradually increases. Equipment failure not only affects production efficiency, but may also cause irreversible pollution to the environment. Therefore, how to timely and accurately monitor the operating status of environmental protection facilities, especially fault diagnosis and emergency treatment, has become a problem that needs to be solved in the current technical field.

[0003] Most of the existing environmental protection facility equipment status monitoring methods rely on traditional physical quantity sensors, such as pressure sensors, flow sensors, etc., but these methods have some limitations. First, traditional sensors can usually only obtain a single type of data, such as temperature, pressure, etc., which are often not comprehensive enough for monitoring complex equipment status and fault types. Secondly, the data processing method in the traditional method is relatively simple, and often determines whether the equipment is in normal working condition by setting thresholds. This method ignores the complex situations and hidden faults that may exist during the operation of the equipment. Therefore, traditional equipment monitoring methods are prone to problems such as underreporting and false alarms, and cannot effectively deal with the changing status and abnormal faults of the equipment.

[0004] In order to overcome the shortcomings of traditional monitoring methods, fusion technology based on multiple sensor information has gradually received attention in recent years. In particular, vibration sensors and infrared temperature sensors are widely used in equipment fault diagnosis. Vibration sensors can monitor the vibration characteristics of equipment during operation, and infrared temperature sensors can reflect the temperature changes on the surface of the equipment. The combination of the two can effectively identify the failure mode of the equipment, such as bearing damage, overheating, friction, etc. However, single sensor data still has certain limitations. Although vibration signals and temperature data can reflect some fault characteristics, their respective effective information volume is relatively limited, and in practical applications they are often affected by factors such as noise and interference, resulting in a decrease in data quality, which in turn affects the accuracy of diagnosis.

[0005] In order to solve the above problems, data fusion methods based on deep learning algorithms have emerged in recent years. By comprehensively analyzing data from multiple sensors, more abundant and comprehensive equipment operation features can be extracted. In particular, spatiotemporal convolutional networks have shown good performance in processing multi-dimensional and time series data. Spatiotemporal convolutional networks can capture the periodic characteristics in vibration data and the spatiotemporal dynamic characteristics in infrared data, effectively improving the accuracy of equipment fault diagnosis. However, the existing equipment fault diagnosis methods based on deep learning still face some challenges, which are mainly reflected in two aspects: First, during the data acquisition process, vibration signals and infrared temperature signals may be interfered by environmental noise and equipment noise, resulting in signal quality degradation. Traditional denoising methods often only use a single filtering algorithm, which is difficult to effectively remove complex interference signals; secondly, existing fault diagnosis systems usually rely on fixed feature extraction methods and lack dynamic adaptability to the equipment operation status. With the continuous changes in the equipment operation status, feature extraction and fault diagnosis models may lose applicability. Therefore, how to adaptively update according to the real-time status of the equipment is another difficult problem facing current technology.

[0006] In addition, the existing technology still has deficiencies in abnormal state analysis. In equipment fault diagnosis, abnormal state detection is a very critical link, especially since early fault modes are often hidden. Traditional abnormality detection methods such as rule-based detection or simple statistical analysis methods are difficult to deal with complex and nonlinear fault modes. Such methods are slow to respond to changes in equipment status, which often results in large losses when the fault is discovered. Therefore, how to combine the historical operation data of the equipment and identify potential abnormal patterns through advanced machine learning and deep learning algorithms is the key to improving fault diagnosis efficiency.

[0007] After processing the fault diagnosis results, how to formulate a suitable processing strategy is also a technical difficulty. Existing control strategies usually rely on fixed rules or manually set parameters for regulation, and lack the ability to dynamically feedback and automatically adjust the real-time status of the equipment. Although traditional control methods can adjust the operating status of the equipment to a certain extent, they often rely on static parameter settings and cannot effectively respond to sudden equipment failures or abnormal conditions. In addition, the maintenance and emergency response processes of the equipment often rely on manual judgment and lack efficient intelligent decision-making support, resulting in long response times and inability to respond in real time.

[0008] Therefore, the equipment status monitoring and fault diagnosis systems in the existing technology generally have problems such as incomplete data collection, unstable data quality, insufficient fault diagnosis capabilities, lack of adaptive update mechanism, and delayed control strategy response. These problems make it difficult for the existing technology to meet the needs of efficient, accurate, and real-time fault diagnosis and emergency treatment in practical applications. Summary of the invention

[0009] One purpose of the present invention is to propose a method for monitoring and diagnosing the status of environmental protection facilities based on vibration and infrared detection. The present invention combines the data of vibration sensors and infrared temperature sensors, uses a spatiotemporal convolutional network to extract the multi-dimensional features of the equipment, and uses an adaptive feature transformation network to achieve data fusion, which can more accurately reflect the operating status of environmental protection facilities and equipment. By using advanced anomaly detection algorithms and fault diagnosis models, it is possible to effectively identify potential faults of equipment and generate appropriate emergency response strategies based on real-time monitoring data, thereby improving the accuracy and response efficiency of equipment fault diagnosis, reducing human intervention, and ensuring the safe operation of environmental protection facilities and equipment.

[0010] According to an embodiment of the present invention, a method for monitoring and diagnosing the state of an environmental protection facility based on vibration and infrared detection comprises the following steps:

[0011] S1. Use vibration sensors and infrared temperature sensors to collect vibration data and temperature data of environmental protection facilities and equipment in real time, perform denoising on the collected vibration data and temperature data, use filtering algorithms to remove noise and environmental interference, standardize the denoised data, and unify the data dimensions;

[0012] S2. Input the processed vibration data and temperature data into the spatiotemporal convolutional network, extract the periodic features in the vibration data and the spatiotemporal dynamic features in the temperature data through the spatiotemporal convolution operation; use the adaptive feature transformation network to deeply fuse the features of the vibration data and the temperature data to generate high-dimensional multimodal features, which reflect the operating status of the environmental protection facilities;

[0013] S3, introduce isolation forest and autoencoder algorithms to detect abnormal patterns of fused multimodal features;

[0014] S4. Based on the results of anomaly detection, a causal inference model is used to analyze the potential causes of abnormal patterns, including equipment failure and overheating. A fault diagnosis report is generated by analyzing current features and historical data. Combined with multimodal feature data, a random forest algorithm is used to classify fault types and further refine the diagnosis results.

[0015] S5. Generate an immediate processing strategy based on the fault type and severity using the proportional integral differential control algorithm, implement the immediate processing strategy in combination with the industrial Internet control system, and control the operating status of environmental protection facilities and equipment in real time;

[0016] S6. Upload the fault diagnosis results and immediate processing strategies to the cloud management platform, and send notifications to relevant maintenance personnel to achieve remote monitoring and collaborative maintenance.

[0017] Optionally, the S1 specifically includes:

[0018] S11. Install vibration sensors at key locations of environmental protection facilities and equipment to collect vibration data of environmental protection facilities and equipment in real time to form a time series data stream, wherein the vibration data is a time series signal representing the vibration amplitude and frequency of the environmental protection facilities and equipment; install infrared temperature sensors on the surface of the equipment to collect temperature data on the surface of the environmental protection facilities and equipment in real time to form a temperature data stream, wherein the temperature data is a time series signal representing the surface temperature of the environmental protection facilities and equipment;

[0019] S12, performing denoising processing on the collected vibration data and temperature data, wherein the denoising processing includes filtering the signal using a filtering algorithm, wherein the filtering algorithm is a Kalman filtering algorithm or a low-pass filtering algorithm, to remove high-frequency noise and environmental interference, and the filtered data are respectively filtered vibration data and temperature data;

[0020] S13, standardize the denoised vibration data and temperature data, and perform mean-variance standardization on the data:

[0021]

[0022] Among them, μ x is the mean value of the vibration data, σ x is the standard deviation of the vibration data, is the filtered vibration data, is the standardized vibration data,

[0023]

[0024] Among them, μ y is the mean of the temperature data, σ y is the standard deviation of the temperature data, is the filtered temperature data, is the standardized temperature data.

[0025] Optionally, the S2 specifically includes:

[0026] S21, inputting the standardized vibration data and temperature data into a spatiotemporal convolutional network, wherein the spatiotemporal convolutional network includes a plurality of spatiotemporal convolutional layers, performing feature extraction on the vibration data and the temperature data through a spatiotemporal convolution operation, and extracting periodic features in the vibration data and spatiotemporal dynamic features in the temperature data;

[0027] S22. In the spatiotemporal convolution layer, the convolution kernel is set to perform convolution operation on the vibration data and the temperature data. The convolution operation calculation process is:

[0028]

[0029] Among them, Zx (t) is the convolution result of vibration data, K x is the convolution kernel size of vibration data, W x (k) is the convolution kernel weight of vibration data, b x is the convolution bias of the vibration data,

[0030]

[0031] Among them, Z y (t) is the convolution result of vibration data, K y is the convolution kernel size of vibration data, W y (k) is the convolution kernel weight of vibration data, b y is the convolution bias for vibration data;

[0032] S23, performing activation function processing on the convolution result and performing down sampling operation;

[0033] S24, fusing the processed vibration data and temperature data with time and space features:

[0034] F(t)=concat(P x (t),P y (t));

[0035] Among them, F(t) is the fused multimodal feature data, P x (t) is the vibration feature data after pooling, P y (t) is the temperature characteristic data after pooling;

[0036] S25. Based on the fusion features output by the spatiotemporal convolutional network, the features of the vibration data and the temperature data are deeply fused using an adaptive feature transformation network, wherein the adaptive feature transformation network includes multiple transformation layers, each of which performs weighted, linear and nonlinear transformations on the input features to extract high-dimensional feature information:

[0037] F transformed (t) = σ(W f ·F(t)+b f );

[0038] Among them, F transformed (t) is the weighted and transformed feature, σ is the activation function, W f is the weight matrix of the transformation layer, b f As the bias term, the adaptive feature transformation network gradually extracts higher-dimensional comprehensive feature information through multi-layer transformation operations:

[0039]

[0040] in, is the output feature of the lth layer transformation, L is the total number of transformation layers, and C(t) is the high-dimensional multimodal feature;

[0041] S26. Regularize the high-dimensional multimodal features, use batch normalization to perform normalization, and use the fully connected layer to perform final fusion and dimensionality reduction on the normalized multimodal features to obtain the final high-dimensional comprehensive features:

[0042]

[0043] Among them, F final (t) is the final high-dimensional comprehensive feature, W c is the weight matrix of the fully connected layer, ∈ is a very small constant to prevent division by zero errors, and b c As the bias term, the adaptive feature transformation network gradually extracts higher-dimensional comprehensive feature information through multi-layer transformation operations.

[0044] Optionally, the S3 specifically includes:

[0045] S31, construct multiple tree structures, calculate the isolation score of each data point by dividing each data point:

[0046]

[0047] Among them, S isolated (c i (t)) is the isolation score of the data point, h(c i (t)) is the average depth of the data point in the isolation forest, α is the hyperparameter of the isolation forest, and c i (t) is a data point;

[0048] S32, judging whether there is an abnormal pattern according to the isolated score, and marking it as an abnormal point when the score exceeds a preset threshold;

[0049] S33. If an abnormal pattern is detected, the autoencoder is used for secondary anomaly detection. The autoencoder compresses the input features into a low-dimensional space, restores them to a high-dimensional output, and evaluates the degree of abnormality of each data point through reconstruction error:

[0050]

[0051] in, is the eigenvalue reconstructed by the autoencoder, ∥·∥2 is the Euclidean distance, E reconstruction (c i (t)) is the reconstruction error;

[0052] S34. After combining the anomaly detection result with other data features, the detected anomaly pattern is output.

[0053] Optionally, the S4 specifically includes:

[0054] S41. Based on the results of anomaly detection, use causal inference models to analyze the potential causes of abnormal patterns;

[0055] S42. Construct a causal graph model to describe the causal relationship between multimodal data. After constructing the causal inference model, use the Bayesian network to infer the causal relationship of each data point and calculate the contribution of each feature to the occurrence of anomalies:

[0056]

[0057] Among them, P(cause(c i (t))|D anomaly (c i (t))) is the causal probability of the feature, D anomaly (c i (t)) is the abnormal data point set, c i (t) is the multimodal feature of the i-th data point, P(D anomaly (c i (t))) is the conditional probability of anomaly detection, P(cause(c i (t))) is the prior probability of the feature;

[0058] S43. Use historical data to establish relationships between features and build regression models to predict potential failures:

[0059] c predict (t) = f(c history (t), parameters);

[0060] Among them, c predict (t) is the predicted current fault feature, c history (t) is the feature in the historical data, f is the regression function, and parameters are the parameters of the regression function;

[0061] S44. Combined with multimodal feature data, the random forest algorithm is used to classify the fault types, the classification results of each tree are calculated, and the final fault type classification is obtained through a voting mechanism:

[0062]

[0063] in, represents the classification result of the i-th tree, N is the number of trees in the random forest, Predict the final fault type.

[0064] S45. Generate a classification result of the fault type through the random forest algorithm, further refine the diagnosis result, and calculate the probability of the refined fault type:

[0065] F fault_type (t) = P(F fault (t)|features);

[0066] Among them, F fault (t) is the predicted fault type, F fault_type (t) is the probability of the fault type under a given feature, and features is a set of multimodal features;

[0067] S46. Output a fault diagnosis report, including the type, severity and potential causes of the fault. The fault types include equipment failure, overheating, bearing damage, and transmission system failure. The report is automatically generated and combined with the operating status data of the environmental protection facilities and equipment to describe the specific location and impact of the fault.

[0068] Optionally, the S5 specifically includes:

[0069] S51. Evaluate severity based on multimodal features and fault classification models:

[0070]

[0071] Among them, F severity (t) is the severity of the fault, w i is the weight of the i-th feature, f i (t) is the value of the i-th feature at time t, and n is the number of features;

[0072] S52, using a proportional-integral-differential control algorithm to generate a control output, wherein the proportional-integral-differential control algorithm adjusts the control input according to the current state of the environmental protection facility equipment:

[0073]

[0074] Among them, U(t) is the control output, K p is the proportional gain, K i is the integral gain, K d is the differential gain, e(t) is the error at the current moment, which is:

[0075] e(t)=F desired (t)-F actual (t);

[0076] Among them, F desired (t) is the expected fault handling target value, F actual (t) is the feedback value of the actual fault processing;

[0077] S53, combining the control output with the fault type, and adjusting the operating status of the environmental protection facilities and equipment through the industrial Internet control system, wherein the adjustment strategy of the control output includes equipment deceleration, emergency shutdown, and switching to a backup system;

[0078] S54. Monitor the operating status of environmental protection facilities and equipment in real time, adjust the parameters in the proportional-integral-differential control algorithm through the feedback mechanism, adapt to the dynamic characteristics of the operating environment and fault changes of environmental protection facilities and equipment, and ensure the adaptability and accuracy of the immediate processing strategy.

[0079] According to an embodiment of the present invention, a system for monitoring and diagnosing the state of environmental protection facilities based on vibration and infrared detection includes:

[0080] Data acquisition module: used to collect vibration data and temperature data of environmental protection facilities and equipment in real time;

[0081] Data preprocessing module: used to denoise the collected vibration data and temperature data, and standardize the denoised data to unify the data dimension;

[0082] Feature extraction module: used to input the processed vibration data and temperature data into the spatiotemporal convolutional network to extract the periodic features in the vibration data and the spatiotemporal dynamic features in the temperature data;

[0083] Feature fusion module: used to input the features of vibration data and temperature data into the adaptive feature transformation network for deep fusion, generate high-dimensional multi-modal comprehensive features, and reflect the operating status of environmental protection facilities;

[0084] Anomaly monitoring module: used to detect abnormal patterns of fused multimodal comprehensive features using isolation forest and autoencoder algorithms;

[0085] Causal model diagnosis module: used to analyze the potential causes of abnormal patterns using causal inference models and generate fault diagnosis reports;

[0086] Fault type classification module: used to classify fault types using random forest algorithm in combination with multimodal data to further refine the diagnosis results;

[0087] Instant processing strategy generation and execution module: used to generate an instant processing strategy based on the fault type and severity using the proportional integral differential control algorithm;

[0088] Implementation module: used to implement immediate processing strategies and control the operating status of environmental protection facilities and equipment in real time, upload fault diagnosis results and processing strategies to the cloud management platform, send notifications to relevant maintenance personnel via SMS, email or APP push, and realize remote monitoring and collaborative maintenance.

[0089] The beneficial effects of the present invention are:

[0090] First of all, the present invention realizes comprehensive monitoring of the operating status of environmental protection facilities and equipment by combining vibration signals with infrared temperature signals. The vibration sensor can capture the mechanical vibration characteristics inside the equipment, while the infrared temperature sensor can reflect the temperature changes on the surface of the equipment. The combination of the two can not only more accurately identify abnormal modes during the operation of the equipment, but also effectively make up for the shortcomings of a single sensor data. Compared with the traditional monitoring method based on a single sensor, the monitoring method provided by the present invention is more comprehensive and has higher accuracy, which can capture small changes in the equipment in real time and discover potential fault hazards in time.

[0091] Secondly, by introducing the spatiotemporal convolutional network technology, the present invention can perform deep learning on the spatiotemporal features in the vibration and temperature signals and extract richer equipment operation features. This deep learning method can not only adapt to the complex operating environment of the equipment, but also capture subtle signals of equipment state changes, greatly improving the accuracy of fault diagnosis. Traditional fault diagnosis methods often rely on fixed feature extraction rules and are difficult to cope with the variability of equipment operating states. The deep learning model of the present invention has adaptive capabilities and can dynamically adjust the diagnostic strategy according to the real-time data of the equipment, thereby improving the flexibility and intelligence level of fault diagnosis.

[0092] In terms of abnormal state analysis, the present invention uses the isolation forest and autoencoder algorithms to effectively avoid the problems of missed reports and false alarms in traditional rule-based detection methods. Fault detection in the prior art often relies on artificially set thresholds or rules, which can easily overlook certain hidden faults or cause delayed responses. Through the detection method of the present invention, abnormal changes in equipment status can be identified in real time, and accurate diagnosis can be performed based on historical operating data, avoiding blind spots in traditional technologies and improving the accuracy and timeliness of fault diagnosis.

[0093] In addition, the instant processing strategy generation and execution module of the present invention enables the system to automatically generate and execute appropriate emergency processing strategies based on the real-time detected equipment fault status. The introduction of this module avoids the drawbacks of relying on manual intervention in traditional systems, making fault processing faster and more accurate, and shortening emergency response time. For those types of faults that require a quick response, the automatic processing capability of the present invention can ensure that the equipment resumes normal operation in a timely manner, reducing equipment downtime, thereby improving production efficiency and facility reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0094] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0095] Figure 1 A flow chart of a method and system for monitoring and diagnosing the state of environmental protection facilities based on vibration and infrared detection proposed by the present invention;

[0096] Figure 2 This is a structural schematic diagram of an environmental protection facility status monitoring and diagnosis method and system based on vibration and infrared detection proposed by the present invention. DETAILED DESCRIPTION

[0097] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0098] refer to Figure 1 and Figure 2 , a method for monitoring and diagnosing the condition of environmental protection facilities based on vibration and infrared detection, comprising the following steps:

[0099] S1. Use vibration sensors and infrared temperature sensors to collect vibration data and temperature data of environmental protection facilities and equipment in real time, perform denoising on the collected vibration data and temperature data, use filtering algorithms to remove noise and environmental interference, standardize the denoised data, and unify the data dimensions;

[0100] S2. Input the processed vibration data and temperature data into the spatiotemporal convolutional network, extract the periodic features in the vibration data and the spatiotemporal dynamic features in the temperature data through the spatiotemporal convolution operation; use the adaptive feature transformation network to deeply fuse the features of the vibration data and the temperature data to generate high-dimensional multimodal features, which reflect the operating status of the environmental protection facilities;

[0101] S3, introduce isolation forest and autoencoder algorithms to detect abnormal patterns of fused multimodal features;

[0102] S4. Based on the results of anomaly detection, a causal inference model is used to analyze the potential causes of abnormal patterns, including equipment failure and overheating. A fault diagnosis report is generated by analyzing current features and historical data. Combined with multimodal feature data, a random forest algorithm is used to classify fault types and further refine the diagnosis results.

[0103] S5. Generate an immediate processing strategy based on the fault type and severity using the proportional integral differential control algorithm, implement the immediate processing strategy in combination with the industrial Internet control system, and control the operating status of environmental protection facilities and equipment in real time;

[0104] S6. Upload the fault diagnosis results and immediate processing strategies to the cloud management platform, and send notifications to relevant maintenance personnel to achieve remote monitoring and collaborative maintenance.

[0105] In this implementation manner, the S1 specifically includes:

[0106] S11. Install vibration sensors at key locations of environmental protection facilities and equipment to collect vibration data of environmental protection facilities and equipment in real time to form a time series data stream, wherein the vibration data is a time series signal representing the vibration amplitude and frequency of the environmental protection facilities and equipment; install infrared temperature sensors on the surface of the equipment to collect temperature data on the surface of the environmental protection facilities and equipment in real time to form a temperature data stream, wherein the temperature data is a time series signal representing the surface temperature of the environmental protection facilities and equipment;

[0107] S12, denoising the collected vibration data and temperature data, wherein the denoising includes filtering the signal using a filtering algorithm, wherein the filtering algorithm is a Kalman filtering algorithm or a low-pass filtering algorithm, to remove high-frequency noise and environmental interference, and the filtered data are respectively filtered vibration data and temperature data:

[0108] S13, standardize the denoised vibration data and temperature data, and perform mean-variance standardization on the data:

[0109]

[0110] Among them, μ x is the mean value of the vibration data, σ x is the standard deviation of the vibration data, is the filtered vibration data, is the standardized vibration data,

[0111]

[0112] Among them, μ y is the mean of the temperature data, σ y is the standard deviation of the temperature data, is the filtered temperature data, is the standardized temperature data.

[0113] In this implementation, S2 specifically includes:

[0114] S21, inputting the standardized vibration data and temperature data into a spatiotemporal convolutional network, wherein the spatiotemporal convolutional network includes a plurality of spatiotemporal convolutional layers, performing feature extraction on the vibration data and the temperature data through a spatiotemporal convolution operation, and extracting periodic features in the vibration data and spatiotemporal dynamic features in the temperature data;

[0115] S22. In the spatiotemporal convolution layer, the convolution kernel is set to perform convolution operation on the vibration data and the temperature data. The convolution operation calculation process is:

[0116]

[0117] Among them, Z x (t) is the convolution result of vibration data, K x is the convolution kernel size of vibration data, W x (k) is the convolution kernel weight of vibration data, b x is the convolution bias of the vibration data,

[0118]

[0119] Among them, Z y (t) is the convolution result of vibration data, K y is the convolution kernel size of vibration data, W y (k) is the convolution kernel weight of vibration data, b y is the convolution bias for vibration data;

[0120] S23, performing activation function processing on the convolution result and performing down sampling operation;

[0121] S24, fusing the processed vibration data and temperature data with time and space features:

[0122] F(t)=concat(P x (t),P y (t));

[0123] Among them, F(t) is the fused multimodal feature data, P x (t) is the vibration feature data after pooling, P y (t) is the temperature characteristic data after pooling;

[0124] S25. Based on the fusion features output by the spatiotemporal convolutional network, the features of the vibration data and the temperature data are deeply fused using an adaptive feature transformation network, wherein the adaptive feature transformation network includes multiple transformation layers, each of which performs weighted, linear and nonlinear transformations on the input features to extract high-dimensional feature information:

[0125] F transformed (t) = σ(W f ·F(t)+b f );

[0126] Among them, F transformed (t) is the weighted and transformed feature, σ is the activation function, W f is the weight matrix of the transformation layer, b f As the bias term, the adaptive feature transformation network gradually extracts higher-dimensional comprehensive feature information through multi-layer transformation operations:

[0127]

[0128] in, is the output feature of the lth layer transformation, L is the total number of transformation layers, and C(t) is the high-dimensional multimodal feature;

[0129] S26. Regularize the high-dimensional multimodal features, use batch normalization to perform normalization, and use the fully connected layer to perform final fusion and dimensionality reduction on the normalized multimodal features to obtain the final high-dimensional comprehensive features:

[0130]

[0131] Among them, F final (t) is the final high-dimensional comprehensive feature, W c is the weight matrix of the fully connected layer, ∈ is a very small constant to prevent division by zero errors, and b c As the bias term, the adaptive feature transformation network gradually extracts higher-dimensional comprehensive feature information through multi-layer transformation operations.

[0132] In this implementation, S3 specifically includes:

[0133] S31, construct multiple tree structures, calculate the isolation score of each data point by dividing each data point:

[0134]

[0135] Among them, S isolated (c i (t)) is the isolation score of the data point, h(c i (t)) is the average depth of the data point in the isolation forest, α is the hyperparameter of the isolation forest, and c i (t) is a data point;

[0136] S32, judging whether there is an abnormal pattern according to the isolated score, and marking it as an abnormal point when the score exceeds a preset threshold;

[0137] S33. If an abnormal pattern is detected, the autoencoder is used for secondary anomaly detection. The autoencoder compresses the input features into a low-dimensional space, restores them to a high-dimensional output, and evaluates the degree of abnormality of each data point through reconstruction error:

[0138]

[0139] in, is the eigenvalue reconstructed by the autoencoder, ∥·∥2 is the Euclidean distance, E reconstruction (c i (t)) is the reconstruction error;

[0140] S34. After combining the anomaly detection result with other data features, the detected anomaly pattern is output.

[0141] In this implementation manner, the S4 specifically includes:

[0142] S41. Based on the results of anomaly detection, use causal inference models to analyze the potential causes of abnormal patterns;

[0143] S42. Construct a causal graph model to describe the causal relationship between multimodal data. After constructing the causal inference model, use the Bayesian network to infer the causal relationship of each data point and calculate the contribution of each feature to the occurrence of anomalies:

[0144]

[0145] Among them, P(cause(c i (t))|D anomaly (c i (t))) is the causal probability of the feature, D anomaly (c i (t)) is the abnormal data point set, c i (t) is the multimodal feature of the i-th data point, P(D anomaly (c i (t))) is the conditional probability of anomaly detection, P(cause(c i (t))) is the prior probability of the feature;

[0146] S43. Use historical data to establish relationships between features and build regression models to predict potential failures:

[0147] c predict (t) = f(c history (t), parameters);

[0148] Among them, c predict (t) is the predicted current fault feature, c history (t) is the feature in the historical data, f is the regression function, and parameters are the parameters of the regression function;

[0149] S44. Combined with multimodal feature data, the random forest algorithm is used to classify the fault types, the classification results of each tree are calculated, and the final fault type classification is obtained through a voting mechanism:

[0150]

[0151] in, represents the classification result of the i-th tree, N is the number of trees in the random forest, Predict the final fault type.

[0152] S45. Generate a classification result of the fault type through the random forest algorithm, further refine the diagnosis result, and calculate the probability of the refined fault type:

[0153] F fault_type (t) = P(F fault (t)|features);

[0154] Among them, F fault (t) is the predicted fault type, F fault_type (t) is the probability of the fault type under a given feature, and features is a set of multimodal features;

[0155] S46. Output a fault diagnosis report, including the type, severity and potential causes of the fault. The fault types include equipment failure, overheating, bearing damage, and transmission system failure. The report is automatically generated and combined with the operating status data of the environmental protection facilities and equipment to describe the specific location and impact of the fault.

[0156] In this implementation manner, S5 specifically includes:

[0157] S51. Evaluate severity based on multimodal features and fault classification models:

[0158]

[0159] Among them, F severity (t) is the severity of the fault, w i is the weight of the i-th feature, f i (t) is the value of the i-th feature at time t, and n is the number of features;

[0160] S52, using a proportional-integral-differential control algorithm to generate a control output, wherein the proportional-integral-differential control algorithm adjusts the control input according to the current state of the environmental protection facility equipment:

[0161]

[0162] Among them, U(t) is the control output, K p is the proportional gain, K i is the integral gain, K d is the differential gain, e(t) is the error at the current moment, which is:

[0163] e(t)=F desired (t)-F actual (t);

[0164] Among them, F desired (t) is the expected fault handling target value, Factual (t) is the feedback value of the actual fault processing;

[0165] S53, combining the control output with the fault type, and adjusting the operating status of the environmental protection facilities and equipment through the industrial Internet control system, wherein the adjustment strategy of the control output includes equipment deceleration, emergency shutdown, and switching to a backup system;

[0166] S54. Monitor the operating status of environmental protection facilities and equipment in real time, adjust the parameters in the proportional-integral-differential control algorithm through the feedback mechanism, adapt to the dynamic characteristics of the operating environment and fault changes of environmental protection facilities and equipment, and ensure the adaptability and accuracy of the immediate processing strategy.

[0167] A condition monitoring and diagnosis system for environmental protection facilities based on vibration and infrared detection, comprising:

[0168] Data acquisition module: used to collect vibration data and temperature data of environmental protection facilities and equipment in real time;

[0169] Data preprocessing module: used to denoise the collected vibration data and temperature data, and standardize the denoised data to unify the data dimension;

[0170] Feature extraction module: used to input the processed vibration data and temperature data into the spatiotemporal convolutional network to extract the periodic features in the vibration data and the spatiotemporal dynamic features in the temperature data;

[0171] Feature fusion module: used to input the features of vibration data and temperature data into the adaptive feature transformation network for deep fusion, generate high-dimensional multi-modal comprehensive features, and reflect the operating status of environmental protection facilities;

[0172] Anomaly monitoring module: used to detect abnormal patterns of fused multimodal comprehensive features using isolation forest and autoencoder algorithms;

[0173] Causal model diagnosis module: used to analyze the potential causes of abnormal patterns using causal inference models and generate fault diagnosis reports;

[0174] Fault type classification module: used to classify fault types using random forest algorithm in combination with multimodal data to further refine the diagnosis results;

[0175] Instant processing strategy generation and execution module: used to generate an instant processing strategy based on the fault type and severity using the proportional integral differential control algorithm;

[0176] Implementation module: used to implement immediate processing strategies and control the operating status of environmental protection facilities and equipment in real time, upload fault diagnosis results and processing strategies to the cloud management platform, send notifications to relevant maintenance personnel via SMS, email or APP push, and realize remote monitoring and collaborative maintenance.

[0177] Embodiment 1:

[0178] The core equipment of a large environmental protection facility is a high-efficiency air purifier. During the operation of the equipment, vibration and temperature changes may be early warning signals of faults. Traditionally, the facility uses a single temperature sensor and vibration sensor to monitor the status of the equipment. However, due to the interference of environmental noise, the noise of the single sensor data is relatively large, and the response to equipment abnormalities is not sensitive enough, resulting in a lag in fault diagnosis and even failure to detect some potential fault problems in time. Based on this situation, the facility decided to implement the fault diagnosis method based on vibration and infrared detection of the present invention, hoping to improve the accuracy and response speed of fault monitoring through multimodal data fusion and deep learning technology. In the implementation of the present invention, multiple vibration sensors and infrared temperature sensors are installed on the environmental protection facility equipment, which are arranged in key parts and on the surface of the equipment respectively.

[0179] In order to verify the effectiveness of the present invention, we conducted a three-month experiment in the application of this environmental protection facility, collected real-time operating data of the equipment, and compared it with traditional monitoring methods.

[0180] Table 1 Performance comparison between the monitoring method of the present invention and the traditional monitoring method

[0181]

[0182]

[0183] In terms of diagnosis time, the monitoring method of the present invention has shown advantages. When dealing with various types of faults, such as bearing wear, overheating faults, temperature sensor faults and motor faults, the traditional monitoring method generally has a long diagnosis time, which can be up to 15 hours. The monitoring method of the present invention greatly shortens the diagnosis time through deep learning models and real-time data processing algorithms. Taking bearing wear faults as an example, the traditional monitoring method takes 10 hours, while the monitoring method of the present invention only takes 1 hour, reducing the diagnosis time by 90%. Similarly, the diagnosis time for overheating faults and temperature sensor faults is also shortened from 6 hours and 12 hours of traditional monitoring methods to 0.5 hours and 1.2 hours, respectively. This improvement means that in practical applications, the monitoring method of the present invention can significantly improve the maintenance efficiency of equipment, reduce equipment downtime, and thus improve overall production efficiency.

[0184] The accuracy of the traditional monitoring method fluctuates greatly when dealing with different faults, with the lowest being only 70% (motor fault), while the average accuracy of the monitoring method of the present invention reaches 93% in all tested fault types. This shows that based on the monitoring method of the present invention, the system can diagnose the fault type more accurately, avoiding the missed diagnosis and misdiagnosis that may be caused by the traditional method, and reducing production interruptions or equipment damage caused by diagnostic errors.

[0185] In traditional monitoring methods, fault diagnosis relies on simple temperature and vibration thresholds. When a device fails, manual intervention is often required to detect and handle it. In the monitoring method of the present invention, the device failure is monitored in real time, and a high-precision fault diagnosis report is generated through data fusion and deep learning algorithms.

[0186] In summary, the monitoring method of the present invention is superior to the traditional monitoring method in terms of efficiency and accuracy of fault diagnosis. Its advantages are not only reflected in the innovation and accuracy of the technology, but more importantly, in the feasibility and efficiency in practical applications. These advantages make the present invention have a wide range of application prospects in the fields of industrial production, equipment maintenance and fault detection, and can significantly improve the intelligent level of equipment maintenance, reduce production costs, and improve the operational efficiency of enterprises.

[0187] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for monitoring and diagnosing the state of environmental protection facilities based on vibration and infrared detection, characterized in that: The steps include: S1. Use vibration sensors and infrared temperature sensors to collect vibration data and temperature data of environmental protection facilities and equipment in real time, perform denoising on the collected vibration data and temperature data, use filtering algorithms to remove noise and environmental interference, standardize the denoised data, and unify the data dimensions; S2. Input the processed vibration data and temperature data into the spatiotemporal convolutional network, extract the periodic features in the vibration data and the spatiotemporal dynamic features in the temperature data through the spatiotemporal convolution operation; use the adaptive feature transformation network to deeply fuse the features of the vibration data and the temperature data to generate high-dimensional multimodal features, which reflect the operating status of the environmental protection facilities; S3, introduce isolation forest and autoencoder algorithms to detect abnormal patterns of fused multimodal features; S4. Based on the results of anomaly detection, a causal inference model is used to analyze the potential causes of abnormal patterns, including equipment failure and overheating. A fault diagnosis report is generated by analyzing current features and historical data. Combined with multimodal feature data, a random forest algorithm is used to classify fault types and further refine the diagnosis results. S5. Generate an immediate processing strategy based on the fault type and severity using the proportional integral differential control algorithm, implement the immediate processing strategy in combination with the industrial Internet control system, and control the operating status of environmental protection facilities and equipment in real time; S6. Upload the fault diagnosis results and immediate processing strategies to the cloud management platform, and send notifications to relevant maintenance personnel to achieve remote monitoring and collaborative maintenance.

2. The environmental protection facility status monitoring and diagnosis method based on vibration and infrared detection according to claim 1 is characterized in that: The S1 specifically includes: S11. Install vibration sensors at key locations of environmental protection facilities and equipment to collect vibration data of environmental protection facilities and equipment in real time to form a time series data stream, wherein the vibration data is a time series signal representing the vibration amplitude and frequency of the environmental protection facilities and equipment; install infrared temperature sensors on the surface of the equipment to collect temperature data on the surface of the environmental protection facilities and equipment in real time to form a temperature data stream, wherein the temperature data is a time series signal representing the surface temperature of the environmental protection facilities and equipment; S12, performing denoising processing on the collected vibration data and temperature data, wherein the denoising processing includes filtering the signal using a filtering algorithm, wherein the filtering algorithm is a Kalman filtering algorithm or a low-pass filtering algorithm, to remove high-frequency noise and environmental interference, and the filtered data are respectively filtered vibration data and temperature data; S13, standardize the denoised vibration data and temperature data, and perform mean-variance standardization on the data: Among them, μ x is the mean value of the vibration data, σ x is the standard deviation of the vibration data, is the filtered vibration data, is the standardized vibration data, Among them, μ y is the mean of the temperature data, σ y is the standard deviation of the temperature data, is the filtered temperature data, is the standardized temperature data.

3. The environmental protection facility status monitoring and diagnosis method based on vibration and infrared detection according to claim 1 is characterized in that: The S2 specifically includes: S21, inputting the standardized vibration data and temperature data into a spatiotemporal convolutional network, wherein the spatiotemporal convolutional network includes a plurality of spatiotemporal convolutional layers, performing feature extraction on the vibration data and the temperature data through a spatiotemporal convolution operation, and extracting periodic features in the vibration data and spatiotemporal dynamic features in the temperature data; S22. In the spatiotemporal convolution layer, the convolution kernel is set to perform convolution operation on the vibration data and the temperature data. The convolution operation calculation process is: Among them, Z x (t) is the convolution result of vibration data, K x is the convolution kernel size of vibration data, W x (k) is the convolution kernel weight of vibration data, b x is the convolution bias of the vibration data, Among them, Z y (t) is the convolution result of vibration data, K y is the convolution kernel size of vibration data, W y (k) is the convolution kernel weight of vibration data, b y is the convolution bias for vibration data; S23, performing activation function processing on the convolution result and performing down sampling operation; S24, fusing the processed vibration data and temperature data with time and space features: F(t)=concat(P x (t),P y (t)); Among them, F(t) is the fused multimodal feature data, P x (t) is the vibration feature data after pooling, P y (t) is the temperature characteristic data after pooling; S25. Based on the fusion features output by the spatiotemporal convolutional network, the features of the vibration data and the temperature data are deeply fused using an adaptive feature transformation network, wherein the adaptive feature transformation network includes multiple transformation layers, each of which performs weighted, linear and nonlinear transformations on the input features to extract high-dimensional feature information: F transformed (t)=σ(W f ·F(t)+b f ); Among them, F transformed (t) is the weighted and transformed feature, σ is the activation function, W f is the weight matrix of the transformation layer, b f As the bias term, the adaptive feature transformation network gradually extracts higher-dimensional comprehensive feature information through multi-layer transformation operations: in, is the output feature of the lth layer transformation, L is the total number of transformation layers, and C(t) is the high-dimensional multimodal feature; S26. Regularize the high-dimensional multimodal features, use batch normalization to perform normalization, and use the fully connected layer to perform final fusion and dimensionality reduction on the normalized multimodal features to obtain the final high-dimensional comprehensive features: Among them, F final (t) is the final high-dimensional comprehensive feature, W c is the weight matrix of the fully connected layer, ∈ is a very small constant to prevent division by zero errors, and b c As the bias term, the adaptive feature transformation network gradually extracts higher-dimensional comprehensive feature information through multi-layer transformation operations.

4. The environmental protection facility status monitoring and diagnosis method based on vibration and infrared detection according to claim 1 is characterized in that: The S3 specifically includes: S31, construct multiple tree structures, calculate the isolation score of each data point by dividing each data point: Among them, S isolated (c i (t)) is the isolation score of the data point, h(c i (t)) is the average depth of the data point in the isolation forest, α is the hyperparameter of the isolation forest, and c i (t) is a data point; S32, judging whether there is an abnormal pattern according to the isolated score, and marking it as an abnormal point when the score exceeds a preset threshold; S33. If an abnormal pattern is detected, the autoencoder is used for secondary anomaly detection. The autoencoder compresses the input features into a low-dimensional space, restores them to a high-dimensional output, and evaluates the degree of abnormality of each data point through reconstruction error: in, is the eigenvalue reconstructed by the autoencoder, ∥·∥2 is the Euclidean distance, E reconstruction (c i (t)) is the reconstruction error; S34. After combining the anomaly detection result with other data features, the detected anomaly pattern is output.

5. The environmental protection facility status monitoring and diagnosis method based on vibration and infrared detection according to claim 1 is characterized in that: The S4 specifically includes: S41. Based on the results of anomaly detection, use causal inference models to analyze the potential causes of abnormal patterns; S42. Construct a causal graph model to describe the causal relationship between multimodal data. After constructing the causal inference model, use the Bayesian network to infer the causal relationship of each data point and calculate the contribution of each feature to the occurrence of anomalies: Among them, P(cause(c i (t))|D anomaly (c i (t))) is the causal probability of the feature, D anomaly (c i (t)) is the abnormal data point set, c i (t) is the multimodal feature of the i-th data point, P(D anomaly (c i (t))) is the conditional probability of anomaly detection, P(cause(c i (t))) is the prior probability of the feature; S43. Use historical data to establish relationships between features and build regression models to predict potential failures: c predict (t)=f(c history (t),parameters); Among them, c predict (t) is the predicted current fault feature, c history (t) is the feature in the historical data, f is the regression function, and parameters are the parameters of the regression function; S44. Combined with multimodal feature data, the random forest algorithm is used to classify the fault types, the classification results of each tree are calculated, and the final fault type classification is obtained through a voting mechanism: in, represents the classification result of the i-th tree, N is the number of trees in the random forest, Predict the final fault type. S45. Generate a classification result of the fault type through the random forest algorithm, further refine the diagnosis result, and calculate the probability of the refined fault type: F fault_type (t)=P(F fault (t)|features); Among them, F fault (t) is the predicted fault type, F fault_type (t) is the probability of the fault type under a given feature, and features is a set of multimodal features; S46. Output a fault diagnosis report, including the type, severity and potential causes of the fault. The fault types include equipment failure, overheating, bearing damage, and transmission system failure. The report is automatically generated and combined with the operating status data of the environmental protection facilities and equipment to describe the specific location and impact of the fault.

6. The environmental protection facility status monitoring and diagnosis method based on vibration and infrared detection as claimed in claim 1 is characterized in that: The S5 specifically includes: S51. Evaluate severity based on multimodal features and fault classification models: Among them, F severity (t) is the severity of the fault, w i is the weight of the i-th feature, f i (t) is the value of the i-th feature at time t, and n is the number of features; S52, using a proportional-integral-differential control algorithm to generate a control output, wherein the proportional-integral-differential control algorithm adjusts the control input according to the current state of the environmental protection facility equipment: Among them, U(t) is the control output, K p is the proportional gain, K i is the integral gain, K d is the differential gain, e(t) is the error at the current moment, which is: e(t)=F desired (t)-F actual (t); Among them, F desired (t) is the expected fault handling target value, F actual (t) is the feedback value of the actual fault processing; S53, combining the control output with the fault type, and adjusting the operating status of the environmental protection facilities and equipment through the industrial Internet control system, wherein the adjustment strategy of the control output includes equipment deceleration, emergency shutdown, and switching to a backup system; S54. Monitor the operating status of environmental protection facilities and equipment in real time, adjust the parameters in the proportional-integral-differential control algorithm through the feedback mechanism, adapt to the dynamic characteristics of the operating environment and fault changes of environmental protection facilities and equipment, and ensure the adaptability and accuracy of the immediate processing strategy.

7. An environmental protection facility status monitoring and diagnosis system based on vibration and infrared detection, which implements the environmental protection facility status monitoring and diagnosis method based on vibration and infrared detection according to any one of claims 1 to 6, characterized in that: include: Data acquisition module: used to collect vibration data and temperature data of environmental protection facilities and equipment in real time; Data preprocessing module: used to denoise the collected vibration data and temperature data, and standardize the denoised data to unify the data dimension; Feature extraction module: used to input the processed vibration data and temperature data into the spatiotemporal convolutional network to extract the periodic features in the vibration data and the spatiotemporal dynamic features in the temperature data; Feature fusion module: used to input the features of vibration data and temperature data into the adaptive feature transformation network for deep fusion, generate high-dimensional multi-modal comprehensive features, and reflect the operating status of environmental protection facilities; Anomaly monitoring module: used to detect abnormal patterns of fused multimodal comprehensive features using isolation forest and autoencoder algorithms; Causal model diagnosis module: used to analyze the potential causes of abnormal patterns using causal inference models and generate fault diagnosis reports; Fault type classification module: used to classify fault types using random forest algorithm in combination with multimodal data to further refine the diagnosis results; Instant processing strategy generation and execution module: used to generate an instant processing strategy based on the fault type and severity using the proportional integral differential control algorithm; Implementation module: used to implement immediate processing strategies and control the operating status of environmental protection facilities and equipment in real time, upload fault diagnosis results and processing strategies to the cloud management platform, send notifications to relevant maintenance personnel via SMS, email or APP push, and realize remote monitoring and collaborative maintenance.

Images