Intelligent fault diagnosis method and device for heat exchange module

Abstract

The invention discloses an intelligent fault diagnosis device of a heat exchange module, and relates to the technical field of fault diagnosis. The system comprises a data acquisition module, a preprocessing module, a feature extraction module, a fault diagnosis model module and a result output module, wherein the data acquisition module accurately acquires operation data at key parts by using a multi-sensor network; the preprocessing module removes noise and abnormal values, smooths and normalizes data, and the feature extraction module extracts multi-level and multi-dimensional features including time domain, frequency domain and time-frequency domain features; the fault diagnosis model adopts a hybrid model combining a capsule network and a long and short term memory network, and the result output module outputs a result in multiple modes and has three-dimensional visualization and multi-channel notification. Diagnosis is accurate, and false alarm and missing alarm are reduced; diagnosis is efficient, and faults are monitored and found in real time; the system has self-adaption and intelligent learning capabilities, adapts to different working conditions, provides convenience for operation and maintenance through a multi-modal output result, has a fault prediction function, and guarantees stable operation of equipment.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and particularly to an intelligent fault diagnosis method and device for a heat exchange module. Background Art

[0002] In industrial production and daily life, heat exchange modules are widely used in fields such as air conditioners, refrigeration equipment, and power systems, and their stable operation is crucial. The background art will be elaborated starting from the deficiencies of the prior art, and then the improvement effects brought by this patent will be described.

[0003] As the core component for realizing heat exchange, heat exchange modules play a key role in many fields such as industrial production, building heating and ventilation, and transportation. In modern industrial production, heat recovery in chemical processes, steam condensation in power plants, etc. all rely on heat exchange modules; in the civilian field, equipment such as air conditioning systems and water heaters also cannot do without them. With the continuous improvement of the requirements for energy utilization efficiency and equipment stability in various industries, the performance and reliability of heat exchange modules directly affect the operating efficiency, energy consumption level, and product quality of the entire system. Once a heat exchange module fails, it will not only lead to energy waste and a decrease in production efficiency, but may also cause equipment damage or even safety accidents in severe cases. Therefore, accurate and timely fault diagnosis of heat exchange modules has become a necessary means to ensure their normal operation.

[0004] Currently, the fault diagnosis of heat exchange modules mainly relies on traditional methods and simple automation technologies. Traditional methods such as manual inspections rely on the experience of technicians to judge the equipment status by observing, touching, listening to sounds, etc. This method is inefficient, highly subjective, difficult to detect early potential faults and internal subtle problems, and cannot achieve real-time monitoring. Existing automated diagnosis technologies mostly rely on single-sensor data or simple threshold judgments. For example, only by monitoring temperature or pressure data and issuing an alarm when the value exceeds the set threshold. These methods ignore the correlation between multiple parameters during the operation of heat exchange modules, have a low diagnostic accuracy for complex faults, cannot distinguish the characteristics of different types of faults, and are prone to false alarms and missed alarms. In addition, existing fault diagnosis systems lack intelligent learning and adaptive capabilities, are difficult to adapt to different working conditions and environmental changes, and cannot self-optimize as the operating state of the equipment changes and over time.

[0005] With the development of industrial intelligence, traditional fault diagnosis technologies have become difficult to meet the increasing operation management requirements of heat exchange modules. On the one hand, the structures and operating conditions of modern heat exchange modules are becoming increasingly complex, posing higher requirements for the accuracy and timeliness of fault diagnosis; on the other hand, with the popularization of the concept of equipment whole-life cycle management, the fault diagnosis system is required to have functions such as fault prediction and intelligent analysis. Therefore, there is an urgent need to develop an intelligent fault diagnosis method and device for heat exchange modules with multi-parameter fusion analysis, intelligent learning, and adaptive capabilities to address the deficiencies of existing technologies and improve the operation reliability and management level of heat exchange modules. Summary of the Invention

[0006] The intelligent fault diagnosis method and device for heat exchange modules proposed by the present invention are used to solve the problems mentioned in the above existing technologies.

[0007] To achieve the above objectives, the present invention adopts the following technical solutions: An intelligent fault diagnosis device for a heat exchange module, comprising: Data acquisition module: A distributed sensor network is used to layout different types of sensors at the inlets and outlets of the heat exchanger, pump body, and valves of the heat exchange module to obtain fluid flow in real time; each sensor transmits the collected temperature, pressure, flow rate, and vibration frequency to the data acquisition terminal through a wired or wireless communication protocol; the acquisition terminal is equipped with a data caching function to handle communication failures; Data preprocessing module: After receiving the collected original operation data, first use the median filtering algorithm to remove noise and isolated outliers; for continuous abnormal data segments, adopt an anomaly detection method based on statistical analysis, and by calculating the interquartile range IQR of the data, determine and correct the data outside the and range as abnormal; then use the adaptive moving average filtering algorithm to smooth the data, and the algorithm automatically adjusts the size N of the moving window according to the fluctuation of the data, and the formula is , where and are adjustable coefficients, is the standard deviation of the data, is the mean of the data; then use the Z-score normalization method to normalize the data, and the formula is , where x is the original data, is the mean of the data, is the standard deviation of the data; Feature extraction module: Extract features from the preprocessed data, use the principal component analysis PCA method to reduce the dimension of the data, and also use independent component analysis ICA to extract independent component features from the data; ICA separates independent components by maximizing the non-Gaussianity of the data, and the formula is , where \(W\) is the separation matrix, \(s\) is the source signal, and \(J\) is the non-Gaussianity metric function; meanwhile, the time-domain features, frequency-domain features, and time-frequency domain features of the data are extracted; the time-frequency domain feature extraction uses the Hilbert-Huang transform (HHT) to decompose the data into intrinsic mode functions (IMFs), and the instantaneous frequency and amplitude features of each IMF are analyzed; Fault diagnosis model module: A hybrid model combining a deep learning-based capsule network (CapsNet) and a long short-term memory network (LSTM) is used for fault diagnosis; the extracted features are input into the hybrid model. The input layer of the model performs preliminary encoding on the features. The loss function of the model uses a dynamic routing loss function, which combines cross-entropy loss and reconstruction loss. The formula is , where is the class label, is the output vector of the capsule, and are the thresholds, is the weight coefficient of the reconstruction loss, \(x\) is the input data, is the reconstructed data; Result output module: In addition to displaying the results of the fault diagnosis in the form of charts and text on the display screen, a 3D visualization model is also provided to show the location and scope of influence of the fault; meanwhile, the result output module is set with a notification function, and the fault information is sent to the operator via text message and email.

[0008] Furthermore, it also includes: Real-time monitoring module: It monitors the operating status of the heat exchange module in real time, analyzes the collected data using a multi-scale analysis method, decomposes the data into different scale spaces through wavelet decomposition, and sets corresponding warning thresholds at each scale; for the data features at different scales, a fuzzy logic reasoning method is used for comprehensive judgment to determine whether to issue a warning signal; the setting of the warning threshold is based on the statistical analysis of historical data and also takes into account the operating conditions and environmental factors of the heat exchange module.

[0009] Model update module: It is used to update the fault diagnosis model regularly. Using a method based on transfer learning, when there are differences between the newly collected operating data and the model training data, or when new fault types appear, the pre-trained model parameters are used as the initial values and fine-tuned on the new data; meanwhile, an active learning mechanism is introduced to select data samples for manual annotation, and then these annotated data are added to the training set to retrain the model.

[0010] Furthermore, the data acquisition module is equipped with a self-calibration function, which regularly self-calibrates the sensors. By comparing with the standard reference value, the calibration curve is fitted using the least squares method. The formula is , where is the measured value of the sensor, is the standard reference value, and a and b are calibration coefficients; meanwhile, the data acquisition module monitors the health status of the sensor and issues a replacement prompt when a sensor failure is detected.

[0011] Furthermore, the feature extraction module also combines a feature selection method of machine learning to remove redundant and irrelevant features and retain the features that contribute to fault diagnosis, improving the training efficiency of the model.

[0012] Furthermore, the fault diagnosis model module also uses the method of federated learning for model training. Data from heat exchange modules distributed in different geographical locations jointly train a global model through encrypted communication and model parameter exchange without sharing the original data.

[0013] Furthermore, the result output module is also equipped with a fault prediction function. By analyzing historical fault data and current operation data, a time series prediction model is used to predict the future fault type and time, providing early warning and decision support for equipment maintenance and management.

[0014] A method for an intelligent fault diagnosis system applying the heat exchange module described above includes: Data acquisition step: Using a distributed sensor network to collect operation data at the parts of the heat exchange module, including temperature, pressure, flow rate, and vibration frequency. Each sensor transmits the data to the acquisition terminal through a wired or wireless communication protocol. Data preprocessing step: First, use the median filtering algorithm to remove noise and isolated outliers from the collected original operation data, and use an anomaly detection method based on statistical analysis to correct continuous anomaly data segments; then use the adaptive sliding average filtering algorithm for smoothing processing. The formula algorithm automatically adjusts the sliding window size N according to data fluctuations, and then uses the Z-score normalization method to normalize the data. Feature extraction step: Extract features from the preprocessed data, use principal component analysis (PCA) and independent component analysis (ICA) for dimensionality reduction. ICA separates independent components by maximizing the non-Gaussianity of the data; meanwhile, extract time domain features, frequency domain features, and time-frequency domain features. The time-frequency domain features are extracted using Hilbert-Huang transform (HHT). Fault diagnosis step: Input the extracted features into a hybrid model combining a capsule network (CapsNet) based on deep learning and a long short-term memory network (LSTM) for fault diagnosis. The model uses a dynamic routing loss function, which is determined by combining the cross-entropy loss and the reconstruction loss formula; determine the fault type, fault occurrence probability, and fault occurrence location according to the model output. Result output step: The fault diagnosis results are displayed through the display screen charts and texts, and the fault location and influence range are displayed by the 3D visualization model; at the same time, the operator is notified by text message and email.

[0015] Furthermore, it also includes: Real-time monitoring step: During the data collection process, the wavelet decomposition of the data is performed using the multi-scale analysis method, and the warning thresholds are set in different scale spaces; the fuzzy logic reasoning is used to comprehensively judge the data characteristics at different scales, and the warning thresholds are dynamically adjusted according to the operating conditions and environmental factors of the heat exchange module. When the data exceeds the warning threshold, a warning signal is issued.

[0016] Model update step: Regularly evaluate the fault diagnosis model. When the new data is abnormal compared with the training data or a new fault type appears, use the method based on transfer learning to fine-tune using the pre-trained model parameters; introduce an active learning mechanism, select data samples for manual annotation and then add them to the training set to retrain the model.

[0017] Compared with the existing technologies, the beneficial effects of the present invention are: In terms of diagnosis accuracy, multi-dimensional data is collected through a distributed multi-sensor network, combined with multi-level feature extraction technologies, such as using methods like principal component analysis and independent component analysis to mine data features, and then a hybrid model combining a capsule network and a long short-term memory network is used for fault diagnosis, which can accurately identify various complex faults, effectively reduce the false alarm and missed alarm rates, and greatly improve the diagnosis accuracy rate compared with traditional methods.

[0018] In terms of diagnosis efficiency, the data preprocessing module uses an adaptive algorithm to quickly process data, and the fault diagnosis model has efficient feature learning and reasoning capabilities, which can realize the real-time monitoring and rapid diagnosis of the operating state of the heat exchange module, can complete a large amount of data processing and analysis in a short time, and timely discover potential faults, avoiding equipment downtime and production losses caused by untimely discovery of faults.

[0019] The device and method of this patent also have strong adaptive and intelligent learning capabilities. The real-time monitoring module can dynamically adjust the warning threshold according to the operating conditions and environmental changes; the model update module continuously optimizes the fault diagnosis model through transfer learning and active learning mechanisms, enabling it to adapt to different devices and operating conditions. In addition, the result output module presents the diagnosis results in a multi-modal manner, not only providing intuitive chart and text information, but also displaying the fault location through 3D visualization, combined with multi-channel notifications and hierarchical reminders, which is convenient for maintenance personnel to timely grasp the device status. At the same time, the fault prediction function can plan equipment maintenance in advance, reduce maintenance costs, improve the stability and reliability of equipment operation, and bring significant economic benefits and safety guarantees to enterprises. Brief Description of the Drawings

[0020] Figure 1 Schematic block diagram of the intelligent fault diagnosis system for the heat exchange module proposed by the present invention; Figure 2 Schematic block diagram of the intelligent fault diagnosis method for the heat exchange module proposed by the present invention; Figure 3 Schematic block diagram for comparing the diagnostic accuracy rates of different fault types of the intelligent fault diagnosis method for the heat exchange module proposed by the present invention; Figure 4 Schematic block diagram showing the variation of the average fault diagnosis time of the intelligent fault diagnosis method for the heat exchange module proposed by the present invention with the data volume. Detailed implementation manners

[0021] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts shall fall within the protection scope of the present invention.

[0022] In the description of the present invention, it should be understood that the terms "center", "longitudinal", "transverse", "length", "width", "thickness", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", "clockwise", "counterclockwise", etc. indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation to the present invention.

[0023] In addition, the terms "first" and "second" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly specifying the quantity of the indicated technical features. Thus, the features defined with "first" and "second" may explicitly or implicitly include one or more of the described features. In the description of the present invention, "a plurality" means two or more unless otherwise specifically defined. In addition, the terms "installation", "connection", and "connection" should be understood in a broad sense. For example, it may be a fixed connection, a detachable connection, or an integral connection; it may be a mechanical connection or an electrical connection; it may be directly connected or indirectly connected through an intermediate medium, and it may be the communication inside two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to specific situations. Next, the present invention will be further introduced in detail in conjunction with the drawings.

[0024] Refer to Figure 1 andFigure 4 :Specific implementation of an intelligent fault diagnosis method and device for a heat exchange module Data acquisition module: Sensors are installed at the inlets and outlets of the heat exchanger, the circulation pump, and the valve of the heat exchange module respectively. The temperature measurement range of the thermistor temperature sensor is -40°C - 200°C, with an accuracy of ±0.1°C, and data is collected once every 0.01 seconds; the piezoresistive pressure sensor has a measurement range of 0 - 10 MPa, with an accuracy of ±0.05%FS, and 10 times of data are collected per second; the electromagnetic flow sensor has a measurement accuracy of ±0.2%, and 5 times of data are collected per second. The sensors transmit data to the data acquisition terminal through the Modbus protocol. The terminal is equipped with a 128GB solid-state drive, which can cache 72 hours of data. When the communication is interrupted, the terminal automatically stores the data and uploads it later when the communication is restored.

[0025] Data preprocessing module: After the original data enters the preprocessing module, the median filtering algorithm with a window size of 3 is first used to remove isolated noise points. For continuous abnormal data, the quartiles Q1 and Q3 are calculated. According to , the data exceeding and range is regarded as abnormal and corrected. Subsequently, according to the data standard deviation and the mean value , the window size N of the adaptive sliding average filter is calculated according to the formula to smooth the data. Finally, the formula is used to perform Z-score normalization on the data, mapping the data to the standard distribution interval.

[0026] Feature extraction module: For the preprocessed data, principal component analysis (PCA) is first used to calculate the covariance matrix , and the principal components with a cumulative contribution rate of 90% are selected. Then, independent component analysis (ICA) is used to separate independent components by maximizing the non-Gaussianity metric function . At the same time, the kurtosis and skewness in the time domain, the power spectral density and frequency bandwidth in the frequency domain, and the time-frequency domain features are obtained through Hilbert-Huang transform (HHT). The data is decomposed into multiple intrinsic mode functions (IMFs) to analyze the instantaneous frequency and amplitude.

[0027] Fault diagnosis model module: The extracted features are input into a hybrid model of a capsule network (CapsNet) and a long short-term memory network (LSTM). The input layer of the capsule network preliminarily encodes the features, and feature combination is realized through the dynamic routing algorithm. The LSTM part processes time series information through the forget gate, input gate, and output gate. The model training uses an improved dynamic routing loss function , where . The model parameters are optimized through the backpropagation algorithm to realize the diagnosis of fault types, probabilities, and positions.

[0028] Result output module: The diagnostic results display the fault location in a 3D visualization model and are presented in a browser through WebGL technology. Meanwhile, a report containing bar charts and line charts is generated to show the fault probability distribution and time trend. Fault information is notified through multiple channels such as enterprise WeChat, email, and SMS, and severe faults trigger audible and visual alarms. Using a time series prediction model, the fault risk for the next 7 days is predicted based on the data of the past 30 days.

[0029] Real-time monitoring module: Perform 5-layer wavelet decomposition on the collected data. At each scale, set the warning threshold according to the mean and standard deviation of historical data. Adopt fuzzy logic reasoning, use the characteristics of parameters such as temperature, pressure, and flow at different scales as inputs, and make a comprehensive judgment through membership functions and fuzzy rule bases. When the fuzzy reasoning result exceeds the set threshold, an early warning is issued.

[0030] Model update module: Evaluate the model weekly. When the distribution difference between new data and training data exceeds the threshold (such as KL divergence greater than 0.3) or a new fault type appears, use the pre-trained model parameters for transfer learning. Through an active learning algorithm, calculate the uncertainty index (such as entropy value) of data samples, select the top 10% of the samples for manual annotation, and add them to the training set to retrain the model.

[0031] Self-calibration function of the data acquisition module: Calibrate the sensor monthly. Compare the data of the sensor measuring known standard values (such as standard pressure blocks, constant temperature baths) with the standard values , fit the calibration curve by the least squares method , and update the calibration coefficients a and b of the sensor. Meanwhile, monitor the drift of the sensor. When the drift exceeds 10% of the accuracy, prompt for replacement. Optimization of the feature extraction module: Use the recursive feature elimination (RFE) algorithm to remove the feature with the smallest decrease in the model prediction accuracy each time, and repeat this process until the number of remaining features reaches 60% of the original features. Then calculate the feature importance through a random forest, screen out the top 50 features with the highest importance, and construct an optimized feature set.

[0032] Federated learning of the fault diagnosis model module: The data of the heat exchange modules in multiple factories train the model locally, and only upload the model parameter gradients to the central server. The central server aggregates the parameter gradients, updates the global model and then distributes it to each factory. Each factory fine-tunes the model with local data to achieve collaborative training under data privacy protection.

[0033] II. Method steps Specific implementation Data acquisition steps: According to the deployment method of the above data acquisition module, install sensors at key parts of the heat exchange module, set the acquisition frequency and communication protocol of the sensors, and transmit the data to the acquisition terminal and cache 72 hours of data. Data preprocessing steps: successively perform median filtering, outlier correction, adaptive moving average filtering and smoothing on the collected raw data, and finally perform Z-score normalization processing to obtain the preprocessed data.

[0034] Feature extraction steps: From the preprocessed data, use PCA and ICA for dimensionality reduction, extract time domain, frequency domain and time-frequency domain features, and optimize the feature set through RFE and random forest algorithms.

[0035] Fault diagnosis steps: Input the optimized features into the CapsNet and LSTM hybrid model, train the model using the improved dynamic routing loss function, and determine the fault type, probability and location according to the model output.

[0036] Result output steps: Output the diagnosis results in the ways of three-dimensional visualization, chart report, multi-channel notification and hierarchical reminder, and use the time series model for fault prediction.

[0037] Real-time monitoring steps: Perform wavelet decomposition on the collected data, set warning thresholds at different scales, and make comprehensive judgments through fuzzy logic reasoning, and issue warnings when the conditions are met.

[0038] Model update steps: Regularly evaluate the model, and when the update conditions are met, use transfer learning and active learning to update the model parameters to improve the model diagnosis ability.

[0039] III. Beneficial effect data representation and explanation 。

[0040] This patent leads the traditional solution by a large margin in terms of fault diagnosis accuracy. It can be seen from the data that due to single-parameter judgment and simple models, the traditional solution is difficult to identify complex faults, while this patent accurately captures fault features through multi-sensor data fusion, multi-level feature extraction and advanced hybrid models. The average fault diagnosis time is significantly shortened because of efficient data processing and fast inference models, which can detect faults in time. The reduction of false alarm rate and missed alarm rate reduces the risks of unnecessary shutdowns and missed faults. The equipment maintenance cost is reduced by 35%, thanks to the early fault prediction and accurate diagnosis, which makes maintenance more planned and avoids the high costs brought by over-maintenance and sudden faults, fully demonstrating the great advantages of this patent in practical applications.

[0041] The above are only the preferred specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, making equivalent substitutions or changes should be covered within the protection scope of the present invention.

Claims

1. An intelligent fault diagnosis device for a heat exchange module, characterized in that: include: Data acquisition module: A distributed sensor network is used to arrange different types of sensors at the heat exchange module inlet and outlet, pump body, and valve to obtain fluid flow in real time; each sensor transmits the collected temperature, pressure, flow, and vibration frequency to the data acquisition terminal through wired or wireless communication protocols; The acquisition terminal is equipped with a data cache function to cope with communication failures; Data preprocessing module: After receiving the collected raw operating data, the median filtering algorithm is first used to remove noise and isolated outliers; For continuous abnormal data segments, an anomaly detection method based on statistical analysis is used to calculate the interquartile range of the data. , will exceed and The data in the range is judged as abnormal and corrected; then the adaptive sliding average filtering algorithm is used to smooth the data. The algorithm automatically adjusts the size N of the sliding window according to the fluctuation of the data. The formula is ,in and is the adjustable coefficient, is the standard deviation of the data, is the mean of the data; then the Z-score normalization method is used to normalize the data, and the formula is , where x is the original data, is the mean of the data, is the standard deviation of the data; Feature extraction module: extract features from preprocessed data, use principal component analysis (PCA) to reduce the dimension of data, and use independent component analysis (ICA) to extract independent component features from data; ICA separates independent components by maximizing the non-Gaussianity of the data. The formula is: , where W is the separation matrix, s is the source signal, and J is the non-Gaussian measurement function; the time domain features, frequency domain features, and time-frequency domain features of the data are extracted simultaneously; The Hilbert-Huang transform (HHT) is used to extract features in the time-frequency domain, decomposing the data into intrinsic mode functions (IMs) and analyzing the instantaneous frequency and amplitude characteristics of each IMF; Fault diagnosis model module: A hybrid model combining the deep learning-based capsule network CapsNet and the long short-term memory network LSTM is used for fault diagnosis; the extracted features are input into the hybrid model, and the input layer of the model performs preliminary encoding on the features. The loss function of the model uses a dynamic routing loss function, combined with cross entropy loss and reconstruction loss, and the formula is: ,in is the category label, is the output vector of the capsule, and is the threshold value, is the weight coefficient of reconstruction loss, For input data, To reconstruct data; Result output module: In addition to displaying the results of fault diagnosis in the form of charts and text on the display screen, it also provides a three-dimensional visual model to display the location and impact range of the fault; at the same time, the result output module is equipped with a notification function, and the fault information is sent to the operator via SMS and email.

2. The intelligent fault diagnosis device for a heat exchange module according to claim 1, characterized in that: Also includes: Real-time monitoring module: monitors the operating status of the heat exchange module in real time, uses multi-scale analysis methods to analyze the collected data, decomposes the data into different scale spaces through wavelet decomposition, and sets corresponding warning thresholds at each scale; for data features at different scales, uses fuzzy logic reasoning methods to comprehensively judge whether it is necessary to issue a warning signal; the setting of the warning threshold is based on statistical analysis of historical data, and also takes into account the operating conditions and environmental factors of the heat exchange module.

3. The intelligent fault diagnosis device for a heat exchange module according to claim 1, characterized in that: Also includes: Model update module: It is used to regularly update the fault diagnosis model. It adopts a transfer learning-based method. When there is a difference between the newly collected operating data and the model training data, or when a new fault type occurs, the pre-trained model parameters are used as the initial values ​​and fine-tuned on the new data. At the same time, an active learning mechanism is introduced to select data samples for manual labeling, and then the labeled data is added to the training set to retrain the model.

4. The intelligent fault diagnosis device for a heat exchange module according to claim 1, characterized in that: The data acquisition module is equipped with a self-calibration function, which periodically performs self-calibration on the sensor. By comparing with the standard reference value, the least squares method is used to fit the calibration curve. The formula is: ,in is the measured value of the sensor, is the standard reference value, a and b are the calibration coefficients; at the same time, the data acquisition module monitors the health status of the sensor and issues a replacement prompt when a sensor failure is detected.

5. The intelligent fault diagnosis device for a heat exchange module according to claim 1, characterized in that: The feature extraction module also combines the feature selection method of machine learning to remove redundant and irrelevant features, retain the features that contribute to fault diagnosis, and improve the training efficiency of the model.

6. The intelligent fault diagnosis device for a heat exchange module according to claim 1, characterized in that: The fault diagnosis model module also uses a federated learning approach to perform model training. Data from heat exchange modules distributed in different geographical locations jointly train a global model through encrypted communication and model parameter exchange without sharing original data.

7. The intelligent fault diagnosis device for a heat exchange module according to claim 1, characterized in that: The result output module is also equipped with a fault prediction function. By analyzing historical fault data and current operation data, a time series prediction model is used to predict the type and time of future faults, providing early warning and decision support for equipment maintenance and management.

8. A method for using the intelligent fault diagnosis device for a heat exchange module according to any one of claims 1 to 7, characterized in that: include: Data collection steps: Use a distributed sensor network to collect operating data at the location of the heat exchange module, including temperature, pressure, flow, and vibration frequency. Each sensor transmits the data to the collection terminal through wired or wireless communication protocols; Data preprocessing steps: The collected raw running data is first subjected to the median filtering algorithm to remove noise and isolated outliers, and the anomaly detection method based on statistical analysis is used to correct the continuous abnormal data segments; then the adaptive sliding average filtering algorithm is used for smoothing, and the formula algorithm automatically adjusts the sliding window size N according to the data fluctuations, and then the Z-score normalization method is used to normalize the data; Feature extraction steps: extract features from the preprocessed data, use principal component analysis PCA and independent component analysis ICA for dimensionality reduction, ICA separates the independent component formula by maximizing the non-Gaussianity of the data; extract time domain features, frequency domain features and time-frequency domain features at the same time, and the time-frequency domain features are extracted using Hilbert-Huang transform HHT; Fault diagnosis steps: The extracted features are input into a hybrid model based on deep learning capsule network CapsNet and long short-term memory network LSTM for fault diagnosis. The model adopts a dynamic routing loss function, combined with the cross entropy loss and reconstruction loss formulas. The fault type, fault probability and fault location are determined based on the model output. Result output steps: The fault diagnosis results are displayed on the display screen through charts and texts, and the three-dimensional visualization model shows the fault location and impact range; at the same time, the operator is notified via SMS and email.

9. The method of the intelligent fault diagnosis device for heat exchange module according to claim 8, characterized in that: Also includes: Real-time monitoring steps: During the data collection process, a multi-scale analysis method is used to perform wavelet decomposition on the data, and warning thresholds are set at different scales; Fuzzy logic reasoning is used to make a comprehensive judgment on the data features at different scales, and the warning threshold is dynamically adjusted according to the operating conditions and environmental factors of the heat exchange module. When the data exceeds the warning threshold, a warning signal is issued.

10. The method of the intelligent fault diagnosis device for heat exchange module according to claim 8, characterized in that: Also includes: Model update step: Regularly evaluate the fault diagnosis model. When new data is abnormal from the training data or a new fault type appears, use a transfer learning-based method to fine-tune the pre-trained model parameters. An active learning mechanism is introduced to select data samples for manual annotation and then add them to the training set to retrain the model.

Images