Preassembled equipment compartment fault prediction method based on multimode data analysis

Abstract

The invention relates to the field of equipment fault prediction, and discloses a pre-installed equipment cabin fault prediction method based on multi-mode data analysis, which comprises the following steps: installing a plurality of sensors in a pre-installed equipment cabin, the plurality of sensors including a temperature and humidity sensor, a gas concentration sensor, a vibration sensor and an electrical quantity sensor, the various sensors are used for collecting environment data and equipment state data in the equipment cabin; performing data preprocessing based on the environment data and the equipment state data, wherein the data preprocessing comprises data cleaning, standardization processing and missing value filling; through a weighted fusion method, multi-mode data of various sensors are fused to generate unified equipment state feature data. Through multi-mode data fusion, the deep learning model and the comprehensive health score, the early-stage accurate prediction of the equipment fault is realized, the accuracy and the response speed of fault early warning are improved, and the equipment operation and maintenance efficiency is optimized.

Description

Technical Field

[0001] The present invention relates to the field of equipment failure prediction, and in particular to a method for predicting failure of a prefabricated equipment cabin based on multi-mode data analysis. Background Art

[0002] In modern industry and equipment management, traditional equipment maintenance methods usually rely on manual inspection or regular inspection, but this method often has errors and delays, making it difficult to detect potential faults in a timely manner. To solve this problem, advanced fault prediction technology, especially fault warning systems based on multi-modal data analysis and intelligent learning, can identify potential equipment failures in advance.

[0003] Existing equipment fault detection technologies mainly rely on traditional monitoring methods. Traditional methods can detect abnormal conditions of equipment to a certain extent by collecting specific parameters and analyzing their changing trends. However, the advantages of these technologies are that they can realize real-time monitoring of a single parameter of the equipment, and after the equipment fails, they can effectively detect certain common failure modes, thereby providing certain support for equipment maintenance.

[0004] However, existing technologies have some obvious shortcomings. First, fault detection methods that rely on single sensor data usually cannot fully reflect the multi-dimensional health status of the equipment. For example, data such as temperature, humidity, vibration, and electrical quantity may affect each other in different scenarios. Traditional technologies cannot effectively integrate these data, resulting in low accuracy of prediction results. Secondly, existing fault prediction methods often rely only on preset thresholds, which lack flexibility and are difficult to cope with changes in equipment under different working environments and loads. Therefore, these technologies often fail to issue early warnings when the equipment status changes slightly. In addition, traditional systems are unable to conduct in-depth analysis of long-term trends in equipment operation, especially when there are potential faults in the equipment but the performance is not obvious. Problems are often not discovered in time, resulting in an increased risk of equipment failure. Summary of the invention

[0005] In view of the shortcomings of the prior art, the present invention provides a prefabricated equipment compartment fault prediction method based on multi-mode data analysis, which solves the problem of single data source and static threshold setting in the prior art, and avoids the problem that traditional methods cannot accurately capture long-term operation changes and multi-dimensional impacts of equipment.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for predicting faults in a prefabricated equipment compartment based on multi-mode data analysis comprises the following steps: Installing multiple sensors in the prefabricated equipment cabin, the multiple sensors including temperature and humidity sensors, gas concentration sensors, vibration sensors and electrical quantity sensors, the multiple sensors are used to collect environmental data and equipment status data in the equipment cabin; Performing data preprocessing based on environmental data and equipment status data, wherein the data preprocessing includes data cleaning, standardization, and missing value filling; Through the weighted fusion method, the multi-mode data of various sensors are fused to generate unified equipment status feature data; A convolutional neural network is used to extract spatial features from the device state feature data, and a long short-term memory network is used to extract time series features from the device state feature data; Based on the spatial characteristics and time series characteristics of the equipment status feature data, a fault prediction model is trained to determine whether the equipment has failed. Through comprehensive health scoring, multiple sensor data and fault prediction results are combined to determine the health status of the equipment. When the equipment status is lower than the set threshold, a fault warning is triggered.

[0007] Preferably, the multiple sensors include: The temperature and humidity sensor is used to collect temperature and humidity data in the equipment cabin; The gas concentration sensor is used to monitor the concentration of dangerous gases; The vibration sensor is used to monitor the vibration of the equipment cabin; The electrical quantity sensor is used to monitor the current, voltage and power electrical parameters of the equipment compartment in real time.

[0008] Preferably, the data preprocessing includes: Data cleaning, standardization and missing value filling; Data cleaning uses the IQR method to remove abnormal data; Missing value imputation used median interpolation to fill missing data; Data standardization is performed to make the mean of the data equal to 0 and the standard deviation equal to 1.

[0009] Preferably, the weighted fusion method includes The weighted fusion method dynamically adjusts the weight of each sensor data; The fused equipment status feature data is used for subsequent fault prediction model training and equipment health status scoring.

[0010] Preferably, the convolutional neural network comprises: The convolutional neural network processes data through multiple convolutional layers and pooling layers; The features extracted after the processing reflect the abnormal conditions of different parts of the equipment.

[0011] Preferably, the long short-term memory network comprises: The LSTM captures the long-term changes of the device status by processing time series features; The time series characteristics can identify potential failure modes of equipment.

[0012] Preferably, the fault prediction model includes: The training process uses the fused device status feature data and device history data; The fault prediction model predicts whether a device fails based on the matching degree between input data and the fault mode.

[0013] Preferably, the device health status includes: Based on the real-time status data and historical operation data of the equipment cabin, combined with multiple sensor data and the time series characteristics of the equipment cabin operation, the preliminary health score of the equipment cabin is calculated; Dynamically adjust equipment health scores based on feedback from multiple sensor data and fault prediction models; When the equipment health score is lower than the set fault threshold, the equipment fault warning is triggered, and the alarm information is automatically generated and notified to the operation and maintenance personnel.

[0014] Preferably, the comprehensive health score includes: Combine data from different sensors and output from fault prediction models to generate a preliminary health score for the equipment; Dynamically adjust the device health score based on historical and real-time data; When the equipment health score is lower than the set threshold, a fault warning is triggered and the operation and maintenance personnel are notified to take necessary maintenance measures.

[0015] Preferably, the equipment health status score includes: Generate a preliminary health score for the device; Adjust device health scores based on historical data; When the health score is lower than the set threshold, a device failure warning is triggered.

[0016] The present invention provides a method for predicting faults in a prefabricated equipment compartment based on multi-mode data analysis. It has the following beneficial effects: 1. The present invention adopts a fault prediction method based on multi-mode data analysis, combines multiple sensor data such as temperature and humidity, gas concentration, vibration, and electrical quantity, and uses weighted fusion technology to generate unified equipment status feature data, achieving a comprehensive and accurate equipment health monitoring effect. Compared with the fault detection method that only relies on a single data source in traditional technology, the present invention solves the problem of incomplete information and improves the accuracy and reliability of fault prediction.

[0017] 2. The present invention uses a combination of convolutional neural networks and long short-term memory networks to extract spatial features and time series features from the equipment status feature data, significantly improving the prediction ability of the fault prediction model. Compared with traditional methods that only rely on static data or simplified models, the present invention solves the problem of being unable to capture long-term equipment operation trends and local anomalies, and provides a higher-precision fault warning.

[0018] 3. The present invention combines data from different sensors and fault prediction results through a comprehensive health scoring system to comprehensively evaluate the health status of the equipment and trigger a fault warning when the equipment health score is lower than the set threshold. Different from the single threshold setting method in traditional technology, the present invention solves the problem of untimely or missed warning triggering, improves the response speed of equipment operation and maintenance, and reduces downtime caused by faults.

[0019] 4. The present invention achieves early warning of equipment failures by using machine learning algorithms and multi-mode data analysis methods, and improves the intelligent operation and maintenance level of equipment. Compared with the traditional method of relying on manual inspections or regular inspections, the present invention solves the problems of low operation and maintenance efficiency and unpredictable failures, and significantly improves the reliability and economic benefits of equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 It is a schematic diagram of the method flow of the present invention. DETAILED DESCRIPTION

[0021] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0022] Please see attached Figure 1 The embodiment of the present invention provides a method for predicting a prefabricated equipment compartment fault based on multi-mode data analysis, comprising the following steps: S1. Installing multiple sensors in the prefabricated equipment cabin, wherein the multiple sensors include temperature and humidity sensors, gas concentration sensors, vibration sensors, and electrical quantity sensors, and the multiple sensors are used to collect environmental data and equipment status data in the equipment cabin; Installing multiple sensors in the pre-installed equipment cabin is the basis for achieving fault prediction and equipment health monitoring. By arranging multiple different types of sensors in the equipment cabin, the environmental data and equipment status data of the equipment can be collected comprehensively and in real time, thereby providing accurate and comprehensive input data for subsequent fault prediction and health assessment. These sensors include temperature and humidity sensors, gas concentration sensors, vibration sensors, and electrical quantity sensors, which monitor different physical quantities respectively. The collection of sensor data will become a key input for the entire system, and the data will be uploaded to the cloud through technologies such as wireless transmission for further analysis and processing. The reliability, real-time nature, and high precision of the data acquisition module directly affect the accuracy of the fault prediction model and the timeliness of the early warning.

[0023] In this embodiment, a variety of sensors are installed in the pre-installed equipment cabin to comprehensively monitor the status and operating environment of the equipment. The sensors include but are not limited to temperature and humidity sensors, gas concentration sensors, vibration sensors, and electrical quantity sensors. Each sensor has different monitoring functions and collection ranges.

[0024] Specifically, temperature and humidity sensors are used to collect temperature and humidity data in the equipment cabin. Temperature and humidity are important parameters that affect the equipment operating environment. Changes in temperature and humidity in the equipment cabin often have a significant impact on the stability and operating efficiency of the equipment. Therefore, real-time monitoring of these environmental parameters helps to detect signs of abnormal equipment operation in advance.

[0025] Gas concentration sensors are used to monitor the concentration of dangerous gases that may exist in the equipment cabin, especially flammable, explosive or toxic gases (such as hydrogen, carbon dioxide, etc.). Abnormal changes in gas concentration may threaten the equipment and even cause equipment failure or safety accidents. Therefore, the role of gas concentration sensors is crucial, and they can promptly warn of gas leaks or concentration changes in the equipment cabin, thereby ensuring the safe operation of the equipment.

[0026] Vibration sensors are used to monitor vibration conditions in the equipment compartment. The vibration status of the equipment directly reflects the health of its mechanical components. Abnormal vibration is usually a precursor to mechanical failure or loosening. By continuously monitoring the vibration data of the equipment, problems such as wear and looseness of equipment components can be discovered in time, thereby avoiding equipment failure. Vibration sensors are essential for monitoring the condition of mechanical equipment, especially in equipment with high loads or high vibrations.

[0027] Electrical quantity sensors are used to monitor electrical parameters such as current, voltage, and power in the equipment compartment in real time. Electrical quantity sensors can measure electrical characteristics such as power consumption, current and voltage fluctuations of equipment in real time, and can reflect the workload and electrical performance of the equipment. Abnormal changes in electrical quantities may indicate problems such as overload, short circuit, or electrical failure of the equipment. Therefore, the monitoring data of electrical quantity sensors is of great significance for the stable operation and fault prediction of equipment.

[0028] In one possible implementation, the data collected by all sensors will be uploaded to the cloud data center through the wireless communication module. The data center will pre-process, extract features and fuse the data from different sensors, and finally generate the health assessment results of the equipment.

[0029] Preprocessing after data collection is crucial. First, the collected sensor data is cleaned to remove outliers. Common cleaning methods include the IQR method (interquartile range method). This method can effectively detect and remove abnormal data and ensure data quality by calculating the interquartile range of the data.

[0030] Next, missing values ​​are filled. During the operation of the device, some data may be missing due to sensor failure or network problems. To avoid the impact of missing data on the analysis results, this embodiment uses the median interpolation method to fill missing values. The median interpolation method calculates the median of known data to fill missing data and ensure data integrity.

[0031] Finally, all data are standardized, with their mean adjusted to 0 and standard deviation adjusted to 1, to ensure that the data collected by each sensor has the same dimension, which is convenient for subsequent data fusion and model training. The formula for standardization is as follows: ; in, is the standardized data, It is The raw data collected by the sensor, It is The mean of the sensor data, is the standard deviation.

[0032] Data fusion is another key step. The weighted fusion method is used to merge the multi-modal data from different sensors to generate unified equipment status feature data. In the weighted fusion method, the data of each sensor will be assigned different weights according to its quality and importance, and the final generated fused data will become the input of the subsequent fault prediction model. The formula of the weighted fusion method is as follows: ; in, Indicates The data of the sensors, Indicates The weight of each sensor, It is the fused device status feature data. is the number of sensors. ∑ represents the summation symbol, which means Equal to 1 to Equal to Sum all items of .

[0033] Combining all sensor data, the generated unified equipment status feature data can fully reflect the health status of the equipment. These feature data will be input into the subsequent deep learning model to extract spatial features and time series features for fault prediction.

[0034] S2. Perform data preprocessing based on environmental data and equipment status data, wherein the data preprocessing includes data cleaning, standardization, and missing value filling; Data preprocessing based on environmental data and equipment status data is a crucial step. This step ensures that the data collected from multiple sensors can play the greatest role in subsequent analysis and provide accurate and clean data input for the fault prediction model. Data preprocessing not only prepares for data fusion and feature extraction, but also improves data quality, thereby enhancing the stability and reliability of the entire prediction system. This process includes three main steps: data cleaning, standardization, and missing value filling.

[0035] In this embodiment, data cleaning is the first step in the processing process, which aims to remove data that does not meet expectations. Since the sensors in the equipment cabin are affected by environmental changes and the operating status of the equipment, the collected data may contain noise or outliers. If these data are not cleaned, they will affect the subsequent analysis results. A commonly used method for data cleaning is the IQR (interquartile range) method, which can effectively identify and remove extreme values ​​or outliers. In general, the IQR method calculates the first quartile (Q1) and the third quartile (Q3) of the data, and then defines the outlier range according to the following formula: ; in, = - is the interquartile range, and data values ​​smaller than this range are considered outliers and are removed from the data set. : represents the third quartile, that is, the lower 75% of the data values ​​after the data set is sorted from small to large. : Represents the first quartile, which is the lower 25% of the data values ​​after the data set is sorted from small to large, to ensure that the data relied on by subsequent analysis is reliable and true.

[0036] Next, after data cleaning, all sensor data will be standardized. In a multi-sensor system, since the dimensions and data ranges of different sensors may vary significantly, directly using these data as input may result in poor model training results. To avoid this, all data needs to be standardized so that they have the same dimensions and data distribution. Specifically, standardization ensures that the scale of all sensor data is consistent by adjusting the mean of each sensor data to 0 and the standard deviation to 1. The formula for standardization is as follows: ; in, is the standardized data, It is The raw data collected by the sensor, For the The average value of the sensors, The standardized data facilitates data fusion, feature extraction, and subsequent model training, which improves the stability and accuracy of the algorithm.

[0037] In some embodiments, data loss may occur during the data collection process, especially in the case of long-term operation or sensor failure. In order to avoid the impact of missing data on subsequent analysis, a missing value filling method is used. Usually, missing value filling adopts the median interpolation method. In the median interpolation method, the missing values ​​will be replaced by the median in the same data sequence, thereby effectively maintaining the statistical characteristics of the data. This method is simple and effective, especially suitable for filling sensor data in equipment cabins. In general, the formula for missing value filling is as follows: ; in, For the filled data, is the median of the data series.

[0038] In a possible implementation, missing value filling can also be interpolated based on the contextual information of the data. For example, when the device state changes slightly, the missing data points can be estimated by linear interpolation. Through these filling methods, the gaps in the data set can be effectively filled to ensure the completeness and accuracy of the analysis results.

[0039] Data cleaning, standardization and missing value filling are the core steps in the data preprocessing process. Through these three technologies, the quality of the equipment cabin sensor data is ensured. The cleaning process removes outliers, the standardization process ensures the consistency of the data scale, and the missing value filling ensures the integrity of the data. After these preprocessed data, not only can it lay a good foundation for the subsequent multi-modal data fusion and feature extraction, but also can improve the accuracy and robustness of the prediction model.

[0040] S3, through the weighted fusion method, the multi-mode data of multiple sensors are fused to generate unified equipment status feature data; The application of weighted fusion methods plays a vital role, especially after fusing multi-modal data from multiple sensors to generate unified equipment status feature data. This process is not only an important part of data preprocessing and feature extraction, but also the core link for multi-modal data analysis to exert efficient prediction capabilities. Through weighted fusion, data from different types of sensors can be integrated to obtain a more comprehensive and accurate equipment health assessment. Combined with the previous steps and modules, especially the cleaning, standardization and missing value filling in the data preprocessing stage, the equipment status feature data generated in this step has high-quality input data, which in turn supports subsequent fault prediction and health assessment tasks.

[0041] In this embodiment, the weighted fusion method is used to fuse multi-mode data from multiple sensors into unified device status feature data. When performing multi-mode data fusion, it is first necessary to ensure the reliability and validity of each sensor data. Through the aforementioned data preprocessing process, including data cleaning and standardization, all sensor data have been converted into the same scale and standard format, providing a basis for fusion. On this basis, the weighted fusion method is used to combine the data of different sensors into a unified feature representation.

[0042] Specifically, the weighted fusion method reflects the importance of different sensor data to the device status by assigning a weight to each sensor data. These weights are automatically learned by the model during the training process and are usually adjusted dynamically based on the quality of the sensor data and its relevance to fault prediction. The weight values ​​are set so that certain key sensor data (such as vibration data or electrical quantity data) account for a larger proportion in the final fusion result, while other sensors (such as ambient temperature and humidity) are assigned appropriate weights based on their role in device health assessment.

[0043] In this way, the generated equipment status characteristic data can be expressed by the following formula: ; in, : represents the integrated equipment status data after fusion, which is the equipment health status characteristic data obtained by weighted summation of all sensor data. :Indicates the The data collected by each sensor can be various parameters such as temperature, pressure, vibration, etc. :Indicates the The weight coefficient of the sensor data indicates the importance of the sensor in the comprehensive state assessment, which is usually determined by model training. : Indicates the total number of sensors, that is, the number of all sensors involved in data collection in the device: In one possible implementation, when some sensor data is of poor quality or cannot be effectively collected, the weight of the sensor will be reduced or set to zero accordingly to ensure the validity of the final device status feature data. This can avoid the adverse impact of a sensor failure or abnormal data on the overall result and improve the robustness of the entire data fusion process.

[0044] The advantage of the weighted fusion method is that it can maximize the advantages of all sensor data, especially when some sensor data is scarce or limited, the data of other sensors can make up for its deficiencies, thereby improving the accuracy of the entire equipment health assessment. In the generated unified equipment status feature data, the data of each sensor will fully express its role in fault prediction in a weighted manner. Ultimately, the fused data is not only more comprehensive, but also more adaptable to different fault types and equipment operating environments.

[0045] Through these steps, the generated equipment status feature data will play a key role in the subsequent fault prediction and health assessment modules. The weighted fusion data provides accurate input for deep learning algorithms and machine learning models, further improving the early prediction capabilities of equipment failures and the accuracy of health status assessment.

[0046] S4, using a convolutional neural network to extract spatial features from the device state feature data, and using a long short-term memory network to extract time series features from the device state feature data; The process of extracting device status feature data is one of the key steps in fault prediction and health assessment. By using deep learning technology, especially convolutional neural networks (CNN) and long short-term memory networks (LSTM), more accurate spatial features and time series features can be extracted from device status feature data. Convolutional neural networks can efficiently capture spatial features, while long short-term memory networks are suitable for processing time series data of equipment, thereby providing rich input information for fault prediction models. These features will be further used to train fault prediction models and help the system make more accurate equipment health assessments.

[0047] In this embodiment, a convolutional neural network (CNN) is used to extract spatial features from device state feature data. Specifically, CNN can automatically extract representations with spatial features from the original device state feature data through multiple convolutional layers and pooling layers. These spatial features reflect abnormal patterns in different parts or dimensions of the equipment, such as the spatial relationship between vibration sensors, temperature and humidity sensors, and electrical quantity sensors. The processing process of the convolutional neural network includes a series of convolution operations, and each convolution layer uses a convolution kernel to filter the input data to extract local features. The pooling layer is used to reduce the dimension of the data and retain important feature information.

[0048] Specifically, the convolution operation can be expressed as: ; in, Represents the input device status characteristic data, Represents the output data after the convolution operation. The feature data obtained after the convolution operation is usually the extracted local spatial features. is the convolution kernel, It is a convolution operation. The convolution process extracts local spatial features by sliding the convolution kernel over the input data. After the convolution, the pooling operation downsamples the data, reducing the amount of calculation while retaining important spatial feature information.

[0049] For example, when the equipment status feature data comes from a vibration sensor, the convolutional neural network can automatically identify possible equipment damage or abnormal vibration patterns by learning the local spatial characteristics of the vibration data, thereby providing support for subsequent fault prediction.

[0050] As an option, the long short-term memory network (LSTM) is used to extract time series features from the device state feature data. Time series features reflect the state changes and trends of the device during long-term operation and can help identify potential failure modes. LSTM is a special recurrent neural network (RNN) that effectively solves the gradient vanishing problem of traditional RNN when dealing with long-term dependencies by introducing a gating mechanism. During the operation of the device, some faults may accumulate gradually, and LSTM can capture such long-term state changes.

[0051] Specifically, the calculation process of the LSTM model is as follows: Forget Gate: ; Input Gate: ; Candidate memory cells: ; Update cell status: ; Output Gate: ; Output: ; in, : Indicates the output of the input gate. Controls the current input data Current cell status The impact of : represents the hidden state at the previous moment, that is, the output value at the previous moment, : represents the output of the output gate, which controls the influence of the current cell state on the hidden state. : represents the hyperbolic tangent activation function, which is used to compress the data between -1 and 1. : Represents the output of the forget gate. Controls the cell state at the previous moment The degree of retention, represents the sigmoid activation function, represents the hyperbolic tangent activation function, , , , is the weight matrix, , , , is the bias term, For the moment The hidden state of For the moment The cell state, : represents the candidate cell state, and updates the cell state at the current moment through the input gate. Through the LSTM network, the long-term dependency of the device state can be captured based on historical data, and possible failure modes can be identified.

[0052] In one possible implementation, LSTM can identify regular changes in the operation of equipment from the time series data collected by sensors, and learn the normal operation mode of the equipment through training. When the equipment status is abnormal, the LSTM network can capture these time series features and promptly warn of potential failures.

[0053] Combining the capabilities of CNN and LSTM, the present invention extracts spatial features from the equipment status feature data through convolutional neural networks, and further uses long short-term memory networks to extract time series features from the same data. These two methods complement each other and can not only capture the spatial anomalies of the equipment, but also analyze the long-term operation trends of the equipment. Through such a feature extraction process, the equipment failure prediction model will obtain more comprehensive data input, thereby improving the prediction accuracy and reliability.

[0054] S5. Based on the spatial characteristics and time series characteristics of the equipment status feature data, a fault prediction model is trained to determine whether the equipment fails through the fault prediction model; Training a fault prediction model based on the spatial and temporal characteristics of the equipment status feature data is one of the key steps in the fault prediction system. Through the aforementioned data processing, feature extraction, and multi-modal data fusion, we have obtained a set of spatial and temporal feature data that can fully reflect the health status of the equipment. These feature data will be used as input to train a prediction model that can accurately determine whether the equipment has failed. The training process of the fault prediction model directly affects the accuracy of the subsequent equipment health status assessment and early warning system.

[0055] In this embodiment, the spatial features and time series features are extracted through the aforementioned steps (such as using convolutional neural networks and long short-term memory networks). These features have fully reflected the spatial changes and long-term state trends of the equipment, and have the ability to predict potential equipment failures. When performing fault prediction, the extracted spatial features and time series features are first used as input data, and a machine learning algorithm is applied to train a fault prediction model. The goal of the model is to determine the current health status of the equipment based on these feature data and predict whether the equipment will fail.

[0056] Specifically, fault prediction models are usually trained using supervised learning algorithms, including common classification algorithms such as support vector machines (SVM), random forests (RF), and XGBoost. The goal of the training process is to enable the model to gradually optimize its prediction capabilities by comparing the input data with the historical fault data, thereby determining whether the equipment will fail. The fault prediction model continuously adjusts its internal parameters to improve its sensitivity to changes in equipment status, thereby being able to more accurately identify potential failure modes.

[0057] Generally, the fault prediction model is trained through the following steps: Input data preparation: The input data of the model is the spatial feature and time series feature data extracted from the above features. The spatial feature reflects the status of the equipment in different parts or dimensions, while the time series feature captures the change trend of the equipment during long-term operation.

[0058] Training process: During the training process, the supervised learning algorithm continuously adjusts the weights and biases of the model by optimizing the objective function, so that the model can fit the relationship between historical fault data and normal data as closely as possible. Specifically, the data input during training With the corresponding label The training goal is to minimize a loss function, such as cross entropy loss or mean squared error (MSE).

[0059] Loss function: The commonly used loss function can be expressed as: ; in, is the loss function, It is The true labels of samples, The model is Input The prediction results, is the parameter of the model, ∑ represents the summation symbol, and represents the Equal to 1 to Equal to Sum all items of .

[0060] Model training and optimization: Through the back-propagation algorithm and optimization methods (such as gradient descent), the model parameters are adjusted to minimize the loss function, thereby improving the accuracy of the model in fault prediction tasks.

[0061] As an option, ensemble learning algorithms (such as random forest or XGBoost) can also be used to train fault prediction models. Ensemble learning combines multiple weak learners to form a strong learner, thereby improving the accuracy and robustness of fault prediction. Each base learner is trained independently, and the prediction results of multiple learners are combined through weighted voting or weighted averaging to obtain the final fault prediction result.

[0062] Specifically, each base learner in the ensemble learning is trained on the device status feature data and generates an independent prediction output. The prediction results of all base learners are fused according to a certain weighted rule and finally output a prediction result. The formula for weighted fusion can be expressed as: ; in, is the final prediction result. It is The prediction results of the base learners are Indicates The weight of each sensor, ∑ represents the summation symbol, and represents the weight of Equal to 1 to Equal to Sum all items of .

[0063] In a possible implementation, the output of the fault prediction model is the probability value of the device failure. The probability value is compared with the set threshold to determine whether the device has failed. If the fault probability value output by the fault prediction model exceeds the preset threshold, the system determines that the device is about to fail and triggers the fault warning mechanism.

[0064] For example, the failure probability output by the fault prediction model is , When is the set fault threshold.

[0065] Model evaluation: After the fault prediction model is trained, the performance of the model needs to be evaluated. Generally speaking, model evaluation uses evaluation indicators such as cross-validation, ROC curve and AUC value to judge the prediction ability of the model. Evaluation indicators help evaluate the performance of the model in practical applications and guide further optimization of the model.

[0066] S6. Combine multiple sensor data with fault prediction results through comprehensive health scoring to determine the health status of the equipment. When the equipment status is lower than the set threshold, a fault warning is triggered.

[0067] The application of comprehensive health scores is an important part of the equipment failure prediction system. By combining the data collected by multiple sensors with the fault prediction results to generate a comprehensive health score, the operating status of the equipment can be more comprehensively evaluated. This comprehensive score provides an efficient equipment health assessment tool that can monitor the status of the equipment in real time and trigger fault warnings in time when the equipment status is abnormal. Specifically, the comprehensive health score comprehensively evaluates the health status of the equipment based on the data from different sensors and the results obtained through the aforementioned fault prediction model. When the health score is lower than the set threshold, the system will immediately issue a fault warning to remind the operation and maintenance personnel to perform necessary inspections and maintenance.

[0068] In this embodiment, the calculation process of the comprehensive health score combines the monitoring data of multiple sensors in the equipment cabin with the fault prediction results to generate a unified health score. First, through the previous steps, the sensor data (such as temperature and humidity, gas concentration, vibration and electrical quantity data) has been collected and preprocessed. These preprocessed data are used to obtain unified equipment status feature data through a weighted fusion method. This feature data provides the necessary input information for the health assessment of the equipment.

[0069] Next, the fault prediction model predicts the probability of equipment failure based on the spatial and time series characteristics of the equipment status feature data. This process generates a device failure probability value, which represents the likelihood of equipment failure. By combining sensor data and fault prediction results, the comprehensive health score can more accurately reflect the overall health status of the equipment.

[0070] Preliminary health score: First, the preliminary health score of the device is calculated based on the device status characteristic data (such as sensor data and fault prediction results). The preliminary health score reflects the current status of the device. This score is usually a comprehensive value calculated by a model based on the relationship between sensor data and historical fault data. The preliminary health score can be expressed as: ; in For the initial health score, It is The raw data collected by the sensor, For the The weight of each sensor, is the equipment failure probability calculated by the failure prediction model, is the number of sensors, ∑ represents the summation symbol, and represents the Equal to 1 to Equal to The preliminary health score combines the sensor data and fault prediction results of the equipment.

[0071] Dynamically adjust the health score: Dynamically adjust the initial health score based on the device's operating history data and real-time monitoring data. Specifically, historical data provides background information for the long-term operating status of the device, and real-time data reflects the current operating status of the device. By combining these two types of data, the dynamically adjusted health score can more accurately reflect the health status of the device. For example, when some sensor data of the device deviates from the normal range, the health score will decrease accordingly. The adjusted health score can be expressed as: ; in, is the adjusted health score, For the initial health score, The score increment is adjusted based on historical and real-time data.

[0072] Fault threshold comparison and warning triggering: When the comprehensive health score is lower than the set threshold, the system automatically triggers a fault warning. This process is completed by comparing with the pre-set health score threshold. The setting of the threshold can be determined based on the device type, operating environment, and historical fault data. In some embodiments, the health score threshold is adjusted according to different equipment and environmental conditions. For example, for some important equipment, the health score threshold may be set lower so that potential faults can be detected in advance and measures can be taken. The health score threshold comparison formula can be expressed as: ; in, is the preset health score threshold, This is the adjusted health score. When the comprehensive health score is lower than this threshold, the fault warning mechanism will be triggered and the system will send an alarm message to the operation and maintenance personnel.

[0073] Generally, the fault warning mechanism can send alarms in a variety of ways, including SMS, email, system notification, etc. After receiving the alarm, the operation and maintenance personnel can check the equipment in time and perform necessary maintenance or repairs to prevent more serious equipment failures.

[0074] As an option, the fault warning system can also be integrated with the equipment's operation control system to automatically shut down or adjust the equipment when the warning is triggered to avoid further deterioration of the equipment failure. For example, when the equipment's health score reaches a set threshold, the system can automatically reduce the equipment load or start the backup equipment to mitigate the impact of the equipment failure.

[0075] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting faults in a prefabricated equipment compartment based on multi-mode data analysis, characterized in that: The following steps are involved: Installing multiple sensors in the prefabricated equipment cabin, the multiple sensors including temperature and humidity sensors, gas concentration sensors, vibration sensors and electrical quantity sensors, the multiple sensors are used to collect environmental data and equipment status data in the equipment cabin; Performing data preprocessing based on environmental data and equipment status data, wherein the data preprocessing includes data cleaning, standardization, and missing value filling; Through the weighted fusion method, the multi-mode data of various sensors are fused to generate unified equipment status feature data; Use convolutional neural networks to extract spatial features from device state feature data, and use long short-term memory networks to extract time series features from device state feature data; Based on the spatial characteristics and time series characteristics of the equipment status feature data, a fault prediction model is trained to determine whether the equipment has a fault. Through comprehensive health scoring, multiple sensor data and fault prediction results are combined to determine the health status of the equipment. When the equipment status is lower than the set threshold, a fault warning is triggered.

2. The method for predicting faults in a prefabricated equipment compartment based on multi-mode data analysis according to claim 1, characterized in that: The various sensors include: The temperature and humidity sensor is used to collect temperature and humidity data in the equipment cabin; The gas concentration sensor is used to monitor the concentration of dangerous gases; The vibration sensor is used to monitor the vibration of the equipment cabin; The electrical quantity sensor is used to monitor the current, voltage and power electrical parameters of the equipment compartment in real time.

3. The method for predicting faults in a prefabricated equipment compartment based on multi-mode data analysis according to claim 1, characterized in that: The data preprocessing includes: Data cleaning, standardization and missing value filling; Data cleaning uses the IQR method to remove abnormal data; Missing value imputation used median interpolation to fill missing data; Data standardization is performed to make the mean of the data equal to 0 and the standard deviation equal to 1.

4. The method for predicting faults in a prefabricated equipment compartment based on multi-mode data analysis according to claim 1, characterized in that: The weighted fusion method includes The weighted fusion method dynamically adjusts the weight of each sensor data; The fused equipment status feature data is used for subsequent fault prediction model training and equipment health status scoring.

5. The method for predicting faults in a prefabricated equipment compartment based on multi-mode data analysis according to claim 1, characterized in that: The convolutional neural network comprises: The convolutional neural network processes data through multiple convolutional layers and pooling layers; The features extracted after the processing reflect the abnormal conditions of different parts of the equipment.

6. The method for predicting faults in a prefabricated equipment compartment based on multi-mode data analysis according to claim 1, characterized in that: The long short-term memory network comprises: The LSTM captures the long-term changes of the device status by processing time series features; The time series characteristics can identify potential failure modes of equipment.

7. The method for predicting faults in a prefabricated equipment compartment based on multi-mode data analysis according to claim 1, characterized in that: The fault prediction model includes: The training process uses the fused device status feature data and device history data; The fault prediction model predicts whether a device fails based on the matching degree between input data and the fault mode.

8. The method for predicting faults in a prefabricated equipment compartment based on multi-mode data analysis according to claim 1, characterized in that: The device health status includes: Based on the real-time status data and historical operation data of the equipment cabin, combined with multiple sensor data and the time series characteristics of the equipment cabin operation, the preliminary health score of the equipment cabin is calculated; Dynamically adjust equipment health scores based on feedback from multiple sensor data and fault prediction models; When the equipment health score is lower than the set fault threshold, the equipment fault warning is triggered, and the alarm information is automatically generated and notified to the operation and maintenance personnel.

9. The method for predicting faults in a prefabricated equipment compartment based on multi-mode data analysis according to claim 1, characterized in that: The comprehensive health score includes: Combine data from different sensors and output from fault prediction models to generate a preliminary health score for the equipment; Dynamically adjust the device health score based on historical and real-time data; When the equipment health score is lower than the set threshold, a fault warning is triggered and the operation and maintenance personnel are notified to take necessary maintenance measures.

10. The method for predicting faults in a prefabricated equipment compartment based on multi-mode data analysis according to claim 4, characterized in that: The device health status score includes: Generate a preliminary health score for the device; Adjust device health scores based on historical data; When the health score is lower than the set threshold, a device failure warning is triggered.

Images