A predictive maintenance system for aerospace equipment

Abstract

The present invention provides a predictive maintenance system for aerospace equipment, comprising: a sensing system for collecting operating parameters of aerospace equipment and transmitting them to a processing and analysis subsystem; a processing and analysis subsystem for performing data processing on the operating parameters of aerospace equipment to obtain characteristic data reflecting the state of aerospace equipment; a fault prediction and diagnosis subsystem for performing fault diagnosis and analysis on the characteristic data reflecting the state of aerospace equipment to obtain the most likely fault type of aerospace equipment at the current moment and the probability of its occurrence; a prediction of aerospace equipment fault trends within a preset time period in the future to obtain the fault type of aerospace equipment within the preset time period in the future and the probability of its occurrence; a health and life prediction subsystem for evaluating and predicting the health status and life of aerospace equipment to obtain a health rating and life curve of aerospace equipment; and a visualization system for displaying process data and processing results of the processing and analysis subsystem, the fault prediction and diagnosis subsystem, and the health and life prediction subsystem.

Description

Technical Field

[0001] The present invention relates to a novel predictive maintenance system for aerospace equipment, which is applicable to the maintenance and guarantee of ground equipment and facilities equipment at major aerospace launch sites. Background Art

[0002] Currently, my country's space launch sites primarily utilize a combination of planned and post-event maintenance. In addition, some key ground equipment utilizes predictive maintenance strategies based on known physical degradation models. Simply opting for post-event maintenance leads to delayed maintenance, and delaying maintenance until equipment failure occurs can easily lead to serious accidents and significant economic losses. Space launches place high demands on system equipment reliability, so preventive maintenance strategies are also incorporated into mission equipment. However, this periodic maintenance strategy, which fails to consider the actual health of the equipment, can lead to over-maintenance and increase mission costs.

[0003] The current aerospace equipment fault maintenance system lacks an integrated, full-process predictive maintenance system from perception to decision-making. At the same time, current aerospace equipment mostly adopts a maintenance strategy of planned maintenance and post-maintenance, resulting in high maintenance costs, and post-maintenance often affects the success rate of space mission execution. Summary of the Invention

[0004] The technical problem solved by the present invention is: to overcome the shortcomings of the existing technology, propose a predictive maintenance system for typical aerospace launch equipment, accurately predict the maintenance of aerospace equipment and provide reasonable maintenance strategies, eliminate the lag in the maintenance of existing typical aerospace equipment, and reduce the cost of aerospace equipment maintenance.

[0005] The technical solution of the present invention is: a predictive maintenance system for aerospace equipment, which includes a sensing subsystem, a processing and analysis subsystem, a fault prediction and diagnosis subsystem, a health and life prediction subsystem, and a visualization subsystem;

[0006] The sensing system collects the operating parameters of aerospace equipment and transmits them to the processing and analysis subsystem;

[0007] The processing and analysis subsystem processes the operating parameters of the aerospace equipment to obtain characteristic data reflecting the status of the aerospace equipment, stores the characteristic data reflecting the status of the aerospace equipment in the parameter database, and sends it to the fault prediction and diagnosis subsystem and the health and life prediction subsystem at the same time;

[0008] The fault prediction and diagnosis subsystem performs fault diagnosis and analysis on characteristic data reflecting the status of aerospace equipment to obtain the most likely type of fault and its probability of occurrence at the current moment; predicts the fault trend of aerospace equipment in a preset time period in the future to obtain the type of aerospace equipment fault and its probability of occurrence in the preset time period in the future, and simultaneously sends the type of aerospace equipment fault and its probability of occurrence at the current moment and in the preset time period in the future to the health and life prediction subsystem;

[0009] The health and life prediction subsystem evaluates and predicts the health and life of aerospace equipment based on characteristic data reflecting the equipment's status, the types of aerospace equipment failures at the current moment and within a preset time period in the future, and their probability of occurrence, thereby obtaining a health rating and life curve for the equipment.

[0010] The visualization system displays the process data and processing results of the processing and analysis subsystem, fault prediction and diagnosis subsystem, and health and life prediction subsystem in a visual manner.

[0011] Preferably, the above-mentioned aerospace equipment predictive maintenance system further includes a support decision management optimization subsystem;

[0012] Ensure the decision-making management optimization subsystem and obtain the processing results of the processing and analysis subsystem, that is, the characteristic data C reflecting the status of aerospace equipment in the past preset time period t , the processing result of the fault prediction and diagnosis subsystem, that is, the fault probability F within the future preset time period t , the processing results of the health and life prediction subsystem, that is, the health rating L in the future preset time period t , forming a multi-source data matrix, multiplying the multi-source data matrix by the input weight matrix for weighting, and using the weighted multi-source data matrix as the input vector The multi-input multi-output LSTM neural network is used to predict the time m when aerospace equipment failure may occur. f , the most cost-effective maintenance time opt , the best maintenance and repair method a, and plan resource management based on the time when aerospace equipment failure may occur, and obtain a form of the demand for the number of spare parts of aerospace equipment within a preset time period.

[0013] Preferably, the multi-input multi-output LSTM neural network comprises an input layer, a hidden layer and an output layer, wherein the input layer tensor is a three-dimensional matrix consisting of the number of samples, the time step, and the sample dimension, and the input vector for:

[0014]

[0015] Among them, w c is the state feature sequence weight, wf is the fault data weight, w l is the weight of healthy life expectancy data; samples is the number of samples;

[0016] There are two hidden layers, denoted as the first hidden layer and the second hidden layer. Each layer consists of 50 neurons. Each neuron contains an input gate, a forget gate, and an output gate. The input gate, forget gate, and output gate are set with different weights.

[0017] The activation function of the output layer adopts the sigmoid function, and the minimization of the mean square error is used as the optimization goal. The result of the hidden layer is mapped to the output and the maintenance decision data H is obtained. t , the maintenance decision data H t for:

[0018]

[0019] in, The time when aerospace equipment failure may occur, For the most cost-effective maintenance time, a t As a means of maintenance.

[0020] Preferably, for mechanical structure aerospace equipment, the state feature sequence weight w c is 0.6, and the healthy life expectancy data w l weight and fault data weight w f 0.2, 0.2 respectively;

[0021] For hydraulic system equipment, the health life data weight w l is 0.6, and the state feature sequence weight w c and fault data weight w f 0.2, 0.2 respectively;

[0022] For control system equipment, the fault data weight w f is 0.6, and the state feature sequence weight w c and healthy life expectancy data w l They are 0.2 and 0.2 respectively.

[0023] Preferably, the multi-input multi-output LSTM neural network is trained by the following method:

[0024] S1. Based on the equipment number tag and time tag, obtain the historical data C reflecting the status characteristics of aerospace equipment t ′, Failure probability historical data F t ′, health rating historical data L t ′, and its corresponding time m when aerospace equipment failure may occur f ′、The most cost-effective maintenance time mopt ', the best maintenance and repair method a';

[0025] S2, will reflect the historical data of aerospace equipment status characteristics C t ′, Failure probability historical data F t ′, health rating historical data L tj Perform input weighting processing respectively to obtain the input vector And the input vector Perform normalization processing to map it to the [0,1] interval;

[0026] S3. Change the time period corresponding to the historical data, repeat steps S1 and S2 to obtain the data set. j=1~J, J is the number of samples;

[0027] S4. Dataset The training set, validation set and test set are divided into 8:1:1 ratios, and the multi-input multi-output LSTM neural network is trained to determine the input gate weight, forget gate weight and output gate weight of each neuron in the hidden layer of the multi-input multi-output LSTM neural network, thereby determining the multi-input multi-output LSTM neural network.

[0028] Preferably, the health and lifespan prediction subsystem includes a health status prediction model and a data-driven RUL model module;

[0029] The health status prediction model module uses a linear regression model, a logistic regression model, or a Gaussian process regression model to fit the health status of aerospace equipment based on historical characteristic data reflecting the status of aerospace equipment. This module obtains a curve showing the health status of aerospace equipment over time and completes the assessment and prediction of the health status of aerospace equipment.

[0030] The data-driven RUL model module uses a Bayesian estimation model to evaluate and predict the remaining service life based on the historical life data of the equipment and the health status change curve over time, obtains the remaining service life curve, and sends the remaining service life curve to the equipment life prediction model module.

[0031] Preferably, the health and life prediction subsystem further includes a performance failure degradation model module;

[0032] The performance failure degradation model is based on the historical characteristic data reflecting the status of aerospace equipment. It uses a linear degradation model or a power degradation model to fit the equipment degradation process, obtain the equipment degradation curve, and predict the equipment degradation trend.

[0033] Preferably, the aerospace equipment operating parameters include equipment performance data, equipment input data, equipment output data, equipment structural data, equipment environment data, and equipment alarm data.

[0034] The beneficial effects of the present invention compared with the prior art are:

[0035] (1) The support decision management optimization subsystem of the present invention adopts a multi-data fusion method based on the LSTM neural network to calculate the optimal time for equipment support, saving equipment maintenance costs and preventing losses caused by equipment downtime. This method is based on predictive maintenance modeling driven by multi-source data fusion and has a certain degree of versatility. It has good adaptability to the diversity and complexity of aerospace equipment. It solves the current problems of delayed planned maintenance of aerospace equipment affecting launch missions, the lack of versatility of general reliability modeling and maintenance, and the lack of intelligent support decision-making and management optimization processes in the current equipment maintenance system, making aerospace equipment maintenance more intelligent and complete.

[0036] (2) The present invention adopts an integrated platform design for aerospace equipment, covering the entire process of aerospace equipment repair and maintenance, including monitoring, processing analysis, fault prediction and diagnosis, health life prediction, result visualization, and guarantee decision-making and management optimization, which greatly facilitates aerospace equipment maintenance management and decision-making. It improves the existing aerospace equipment maintenance system business is relatively scattered and equipment maintenance management technology is not universal;

[0037] (3) The present invention trains the maintenance decision model based on the characteristics of aerospace equipment, equipment status feature data, fault history data, health life history data, and maintenance decision history data, taking into account the comprehensiveness of the maintenance decision training data dimensions and ensuring the accuracy of the optimal maintenance decision. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 This is a block diagram of a predictive maintenance system for aerospace equipment according to an embodiment of the present invention;

[0039] Figure 2 This is a general structural diagram of a predictive maintenance system for aerospace equipment according to an embodiment of the present invention;

[0040] Figure 3 This is a flow chart of the integrated, platform-based predictive maintenance technology system for aerospace equipment according to an embodiment of the present invention;

[0041] Figure 4 This is a dynamic maintenance decision-making process framework for fault and life prediction based on real-time data in an embodiment of the present invention. DETAILED DESCRIPTION

[0042] The present invention will be further described below in conjunction with the embodiments.

[0043] With the development of artificial intelligence and Internet of Things technologies, this invention helps aerospace equipment maintenance shift from "preventive" maintenance to "predictive" maintenance by accumulating equipment's measurement point data, failure mechanism models, and life cycle-related parameters. It is a feasible, more efficient, and more reliable means of equipment maintenance.

[0044] like Figure 1 、 Figure 2 、 Figure 3 As shown, the present invention provides a predictive maintenance system for aerospace equipment, comprising a sensing subsystem 1, a processing and analysis subsystem 3, a fault prediction and diagnosis subsystem 4, a health and life prediction subsystem 5, a visualization subsystem 6, and a support decision management optimization subsystem 7. The system connects to the aerospace equipment's equipment monitoring sensors via the sensing subsystem and transmits information to the processing and analysis subsystem, the fault prediction and diagnosis subsystem, the health and life prediction subsystem, the visualization subsystem, and the support decision management optimization subsystem via a sensing network 2. This system implements monitoring, analysis, fault diagnosis, health and life prediction, visualization of fault results, and maintenance and support decision-making for aerospace equipment faults, thus forming a comprehensive intelligent maintenance and support process for aerospace equipment.

[0045] Data from perception subsystem 1 is transmitted to processing subsystem 3 via perception network 2. Network interfaces connect processing subsystem 3, fault prediction and diagnosis system 4, health and life prediction system 5, maintenance visualization system 6, and support decision management and optimization system 7. Perception network 2 can be a public information network such as the Internet of Things.

[0046] Predictive maintenance perception system 1 utilizes IoT devices to collect relevant measurement point data from aerospace equipment and derives the equipment's operating parameters from this data. The perception system consists of three layers: the perception layer, the network layer, and the application layer. The perception layer collects and monitors measurement point data based on the fault type. Sensors acquire physical measurement data such as deformation, pressure, velocity, temperature, voltage, and current at key measurement points. These signals are converted into digital signals and transmitted via perception network 2 to the aerospace equipment predictive maintenance processing subsystem 3 for processing and storage. Measurement point data includes, but is not limited to, operating status information, fault data, repair and maintenance data, and performance data. After obtaining the aerospace equipment's operating parameters, processing subsystem 3 unifies and aligns the parameter names and dimensions of the different types of equipment. Signal processing and data fusion algorithms are used to filter out clutter and noise signals. Feature extraction and data mining are then used to obtain characteristic data reflecting the equipment's status. After obtaining characteristic data reflecting the status of aerospace equipment, it is stored in a database and transmitted via the network to the Fault Prediction and Diagnosis Subsystem 4. Using signal processing analysis, fault tree diagnosis, expert scoring system fault diagnosis analysis, support vector machines, and time series analysis, fault characteristic analysis and fault trend prediction are performed. The current cause and content of the aerospace equipment fault are determined. While monitoring aerospace fault information, a prediction of the equipment's failure probability within a certain period of time is also provided. The calculation results from the Processing Subsystem 3 and the Fault Prediction and Diagnosis Subsystem 4 are aggregated into the Health and Life Prediction Subsystem 5. The Health and Life Prediction Subsystem calculates the current health and life of the aerospace equipment using a performance failure degradation model, a health status prediction model, and a data-driven RUL model. Based on the calculation results from the Processing Subsystem 3, the Fault Prediction and Diagnosis Subsystem 4, and the Life Prediction Subsystem 5, the Decision Management Optimization Subsystem 7 generates corresponding equipment resource management planning content based on a dynamic maintenance decision optimization model and a reinforcement learning maintenance decision model, and automatically generates a schedule. The visualization content in these systems can be displayed through trend charts and graphs using the visualization BI tools of the Visualization System 6.

[0047] The following is a detailed description of each subsystem:

[0048] 1. Perception subsystem

[0049] The perception system collects aerospace equipment operating parameters and transmits them to the processing and analysis subsystem; the aerospace equipment operating parameters include equipment performance data, equipment input data, equipment output data, equipment structural data, equipment environment data, and equipment alarm data.

[0050] The sensors are connected to the sensing subsystem 1 via ZigBee and CAN bus technologies;

[0051] Equipment performance data: such as speed, pressure, load, temperature, noise, vibration, etc.;

[0052] Equipment input data: such as power supply, hydraulic power, air, etc.;

[0053] Equipment output data: such as power, traction, pressure, etc.;

[0054] Equipment structural data: such as location, material, stiffness, flexibility, fatigue, thermal expansion, etc.;

[0055] Equipment environment data: such as water, wind, temperature, altitude, humidity, etc.;

[0056] Equipment alarm data: such as overspeed, current overload, voltage overvoltage or undervoltage, etc.;

[0057] For the horizontal rod component of the pendulum, the aerospace equipment operating parameters include position, material, stiffness, and vibration signal;

[0058] For the rocker arm hydraulic oil pump equipment, the aerospace equipment operating parameters include the oil pump operating status, oil cylinder operating pressure, swing angle, and swing signal;

[0059] For the rocker arm control cabinet equipment, the aerospace equipment operating parameters include operating temperature, voltage value, and current value.

[0060] For the UPS equipment of the swing arm control cabinet, the operating parameters of the aerospace equipment include operating temperature values, voltage and current values, power values, load values, and capacity values.

[0061] This subsystem establishes the hardware and software measurement points for the perception layer of predictive maintenance. The location of these measurement points is related to the specific aerospace equipment failure mechanism and the input required for predictive maintenance, establishing the input foundation for data collection and model building.

[0062] 2. Processing and analysis subsystem

[0063] There are numerous types of aerospace equipment, and the status data collected from each type of equipment suffers from varying parameter names and dimensions. Directly sampling parameters also makes it difficult to characterize the equipment's status. This system processes the equipment's operating parameters to meet the needs of status monitoring and subsequent diagnosis and evaluation. Therefore, the processing and analysis subsystem processes the equipment's operating parameters to obtain characteristic data reflecting the equipment's status. This data is stored in a parameter database and simultaneously sent to the fault prediction and diagnosis subsystem and the health and life prediction subsystem.

[0064] Data processing includes one or more of feature extraction, signal processing, data mining and data fusion.

[0065] Signal processing methods include: missing value filling, interpolation, outlier processing, noise reduction, data smoothing, data extrapolation, etc.

[0066] Feature extraction methods include: time domain and frequency domain feature extraction, statistical feature extraction, fast Fourier transform, PCA, etc.

[0067] Data mining includes: decision tree algorithm, genetic algorithm, nearest neighbor algorithm, Bayesian network, maximum expectation algorithm, etc.;

[0068] Data fusion includes competitive fusion, complementary fusion, collaborative fusion and other methods.

[0069] The characteristic data reflecting the status of aerospace equipment may be the average value and variance of equipment performance data, the maximum value of equipment output data, the frequency spectrum of equipment environment data, etc.

[0070] For the pendulum horizontal rod component, the characteristic data reflecting the state of aerospace equipment is vibration signal data. By setting a threshold to filter noise information and using Fourier transform to extract and monitor non-stationary signals, characteristic data reflecting the state of aerospace equipment can be obtained.

[0071] For the rocker arm hydraulic oil pump equipment, the corresponding characteristic data reflecting the status of aerospace equipment are the historical data of the oil pump start-up time, the rocker arm swing angle, and the rocker arm in place mark. The mean standard deviation calculation method is used to extract the characteristic data reflecting the status of aerospace equipment.

[0072] For the rocker control cabinet equipment, the corresponding characteristic data reflecting the status of the aerospace equipment are high temperature alarm signs, current overload signs, and voltage overvoltage or low voltage sign data. The characteristic data reflecting the status of the aerospace equipment are extracted by collaborative fusion.

[0073] For the UPS equipment of the swing arm control cabinet, the corresponding characteristic data reflecting the state of the aerospace equipment are voltage, current and power value data. The Bayesian method is used to extract the characteristic data reflecting the state of the aerospace equipment.

[0074] 3. Fault prediction and diagnosis subsystem

[0075] The fault prediction and diagnosis subsystem performs fault diagnosis and analysis on characteristic data reflecting the status of aerospace equipment to obtain the most likely type of fault and its probability of occurrence at the current moment; predicts the fault trend of aerospace equipment in a preset time period in the future to obtain the type of aerospace equipment fault and its probability of occurrence in the preset time period in the future, and simultaneously sends the type of aerospace equipment fault and its probability of occurrence at the current moment and in the preset time period in the future to the health and life prediction subsystem;

[0076] The fault prediction and diagnosis subsystem includes the fault diagnosis subsystem and the fault prediction subsystem.

[0077] The fault diagnosis subsystem adopts signal processing analysis, fault tree diagnosis or expert scoring system fault diagnosis analysis.

[0078] The fault prediction subsystem uses support vector machines and time series analysis to predict the failure trends of aerospace equipment in the future time period.

[0079] The above system utilizes the working status data of the equipment in the healthy, abnormal, and faulty states accumulated during the long-term operation of the equipment, and uses a large amount of sample data in the healthy state to extract the equipment failure characteristics using machine learning methods, build equipment fault classification and identification models, and apply them to the subsequent identification and diagnosis of similar faults.

[0080] 4. Health and life expectancy prediction subsystem

[0081] The health and life prediction subsystem evaluates and predicts the health status and life of aerospace equipment based on characteristic data reflecting the status of aerospace equipment, the types of aerospace equipment failures and their probability of occurrence at the current time and in the future preset time period, and obtains the health rating and life curve of aerospace equipment.

[0082] The health and life prediction subsystem evaluates and predicts the health status and life of the equipment based on performance failure degradation modeling, health status prediction model and data-driven RUL model.

[0083] The health status prediction model module uses linear regression, logistic regression, or Gaussian process regression models to fit the health status of aerospace equipment based on historical characteristic data reflecting the status of aerospace equipment. This module obtains a curve of the health status of aerospace equipment over time and completes the assessment and prediction of the health status of aerospace equipment.

[0084] The data-driven RUL model module uses a Bayesian estimation model to evaluate and predict the remaining useful life (RUL) based on the historical life data of the equipment and the health status change curve over time, obtains the remaining useful life curve, and sends the remaining useful life curve to the equipment life prediction model module.

[0085] The performance failure degradation model is based on the historical characteristic data reflecting the status of aerospace equipment. It uses a linear degradation model or a power degradation model to fit the equipment degradation process, obtain the equipment degradation curve, and predict the equipment degradation trend.

[0086] The linear degradation model is applicable to devices with linear degradation characteristics;

[0087] For devices with nonlinear exponential degradation characteristics, the power degradation model is applicable;

[0088] 5. Ensure decision-making management optimization subsystem

[0089] Predictive maintenance decision-making and management optimization technology is a maintenance decision-making method based on equipment operating status, designed to reduce maintenance costs while ensuring the continued efficient operation of production systems. By continuously monitoring equipment operating status, the system predicts when failures are likely to occur, identifies the most cost-effective maintenance opportunity before a failure occurs, and implements appropriate maintenance activities to prevent deterioration of equipment operating status, minimize unplanned downtime, and reduce maintenance costs.

[0090] like Figure 4 As shown in the figure, the guarantee decision management optimization subsystem obtains the processing results of the processing analysis subsystem, that is, the characteristic data C reflecting the status of aerospace equipment in the past preset time period. t , the processing result of the fault prediction and diagnosis subsystem, that is, the fault probability F within the future preset time period t , the processing results of the health and life prediction subsystem, that is, the health rating L in the future preset time period t , forming a multi-source data matrix, multiplying the multi-source data matrix by the input weight matrix, and taking the weighted multi-source data matrix as the input vector The multi-input multi-output LSTM neural network is used to predict the time m when aerospace equipment failure may occur. f , the most cost-effective maintenance time opt , the best maintenance and repair method a, and plan resource management based on the time when aerospace equipment failure may occur, and obtain a form of the demand for the number of spare parts of aerospace equipment within a preset time period.

[0091] The multi-input multi-output LSTM neural network includes an input layer, a hidden layer and an output layer, wherein the input layer tensor is a three-dimensional matrix consisting of the number of samples, the time step and the sample dimension.

[0092] In a specific embodiment of the present invention, the input layer tensor input = (samples, timesteps, data_dim), which constitutes a three-dimensional matrix of samples*timesteps*data_dim, where the value range of samples is at least greater than 100; the value range of timesteps is 1 to 3, and data_dim is 3.

[0093] Input vector for:

[0094]

[0095] is the multi-source input weight, where w c is the state feature sequence weight, w f is the fault data weight, wl is the weight of healthy life expectancy data, and samples is the number of samples;

[0096] For mechanical structure aerospace equipment (for example, the transmission components of the launch tower swing arm), the training input is mainly based on the equipment state feature data of the time series, and the state feature sequence weight w c is 0.6, the healthy life history data and the failure history data are used as auxiliary inputs, and the healthy life data w l weight and fault data weight w f 0.2, 0.2 respectively;

[0097] For hydraulic system equipment (e.g., hydraulic pumps, relief valves), the training input is mainly based on the time series of healthy life history data. The healthy life data w l The weight is 0.6, the equipment status feature data and fault history data are used as auxiliary inputs, and the status feature sequence weight w c and fault data weight w f 0.2, 0.2 respectively;

[0098] For control system equipment (e.g., control cabinet), the training input is mainly based on the time series fault history data, with the fault data weight being 0.6, and the equipment status feature data and health life history data as auxiliary inputs, with the status feature sequence weight w c and healthy life expectancy data w l They are 0.2 and 0.2 respectively.

[0099] There are two hidden layers, denoted as the first hidden layer and the second hidden layer. Each layer consists of 50 neurons. Each neuron contains an input gate, a forget gate, and an output gate. The input gate, forget gate, and output gate are set with different weights.

[0100] The activation function of the output layer adopts the sigmoid function, and the minimization of the mean square error is used as the optimization goal. The result of the hidden layer is mapped to the output and the maintenance decision data H is obtained. t , the maintenance decision data H t for:

[0101]

[0102] in, The time when aerospace equipment failure may occur, For the most cost-effective maintenance time, a t As a means of maintenance.

[0103] The above multi-input multi-output LSTM neural network is trained by the following method:

[0104] S1. Based on the equipment number tag and time tag, obtain the historical data C reflecting the status characteristics of aerospace equipment t ′, Failure probability historical data F t ′, health rating historical data L t ′, and its corresponding time m when aerospace equipment failure may occur f ′、The most cost-effective maintenance time m opt ', the best maintenance and repair method a';

[0105] S2, will reflect the historical data of aerospace equipment status characteristics C t ′, Failure probability historical data F t ′, health rating historical data L tj Perform input weighting processing respectively to obtain the input vector And the input vector Perform normalization processing to map it to the [0,1] interval;

[0106] S3. Change the time period corresponding to the historical data, repeat steps S1 and S2 to obtain the data set. j=1~J, J is the number of samples;

[0107] S4. Dataset The training set, validation set, and test set are divided into 8:1:1 ratios to train the multi-input multi-output LSTM neural network. The input gate weight, forget gate weight, and output gate weight of each neuron in the hidden layer of the multi-input multi-output LSTM neural network are determined, thereby determining the multi-input multi-output LSTM neural network. The health and lifespan prediction subsystem includes a health status prediction model, a data-driven RUL model module, and a performance failure degradation model module.

[0108] The health status prediction model module uses linear regression, logistic regression, or Gaussian process regression models to fit the health status of aerospace equipment based on historical characteristic data reflecting the status of aerospace equipment. This module obtains a curve of the health status of aerospace equipment over time and completes the assessment and prediction of the health status of aerospace equipment.

[0109] The data-driven RUL model module uses a Bayesian estimation model to evaluate and predict the remaining useful life (RUL) based on the historical life data of the equipment and the health status change curve over time, obtains the remaining useful life curve, and sends the remaining useful life curve to the equipment life prediction model module.

[0110] The performance failure degradation model is based on the historical characteristic data reflecting the status of aerospace equipment. It uses a linear degradation model or a power degradation model to fit the equipment degradation process, obtain the equipment degradation curve, and predict the equipment degradation trend.

[0111] The linear degradation model is applicable to devices with linear degradation characteristics;

[0112] For devices with nonlinear exponential degradation characteristics, the power degradation model is applicable;

[0113] 5. Visualization system

[0114] The visualization system will display the process data and processing results of the processing analysis subsystem, fault prediction and diagnosis subsystem, and health and life prediction subsystem in a visual way.

[0115] Data visualization for aerospace equipment is a crucial tool for predictive maintenance. It helps non-professional maintenance engineers understand equipment status in a timely manner. Using visualization elements such as charts, graphs, and maps, data visualization tools make it easy to view and understand trends, outliers, and patterns in equipment data. This is crucial for analyzing massive amounts of information and making data-driven decisions. The system uses visual comparisons to analyze trends in overall equipment availability, decreasing repair costs, decreasing repair inventory, and post-maintenance trends, as well as analyzing the execution status of spot checks and periodic maintenance. This allows for targeted analysis of frequent failures, long-term downtime, and high-cost failures.

[0116] System dynamics process:

[0117] (1) The sensing subsystem reports the status of relevant measurement points through the sensor network through the equipment sensors and reports it to the processing and analysis system.

[0118] (2) The processing and analysis system performs signal processing, data fusion, and data mining on the data to extract characteristic data reflecting the status of the aerospace equipment, and stores the characteristic data reflecting the status of the aerospace equipment in the database.

[0119] (3) The fault prediction and diagnosis system obtains the corresponding characteristic parameter data, performs fault characteristic analysis on the signal, and predicts the fault trend.

[0120] (4) The health and life prediction system comprehensively processes the characteristic data provided by the analysis system and the fault characteristic analysis and trend prediction data provided by the fault prediction and diagnosis system to predict the life of equipment components.

[0121] (5) Based on the real-time fault trend prediction and life prediction results, the guarantee decision-making and management optimization system gives the corresponding dynamic maintenance decision conclusion and gives a reasonable resource management plan.

[0122] The present invention includes the following key technologies:

[0123] (1) Integrated and platform-based predictive maintenance technology system for aerospace equipment

[0124] This technology realizes the full-process guarantee of aerospace equipment maintenance, including monitoring, analysis and processing, feature extraction, fault feature analysis, fault trend prediction, dynamic maintenance decision-making and resource management planning. It covers the corresponding historical maintenance guarantee, real-time status and predictive maintenance of aerospace equipment. The maintenance of aerospace ground equipment plays a vital role in reducing the frequency of failures, improving operational efficiency and ensuring launch quality. In traditional maintenance methods, fault feature analysis, fault prediction, health status and maintenance management are different subsystems. In the design of this system, the fault analysis is integrated and combined with the current health status of the equipment to give a dynamic fault life prediction, and dynamic maintenance decisions are made based on the prediction results. It improves the utilization rate of data between systems, integrates the scattered system analysis results, reduces the manual analysis cost of equipment maintenance and repair, and improves decision-making power.

[0125] This technology aims to reduce maintenance costs while ensuring continuous and efficient operation of production systems. Specifically, by continuously monitoring equipment operating status, it can predict when failures are likely to occur, identify the most cost-effective maintenance opportunities before they occur, and implement appropriate maintenance activities to prevent equipment degradation, thereby minimizing unplanned downtime and reducing maintenance costs. This platform can effectively reduce maintenance costs and ensure mission success rates.

[0126] (2) Dynamic maintenance decision-making technology based on fault and life prediction of real-time data

[0127] During the operation of aerospace equipment, lifespan predictions and maintenance decisions derived from real-time data at each moment can be extrapolated to predict future lifespan predictions and maintenance decisions. This technology primarily divides time intervals into distinct time intervals. Based on the current time interval status, it dynamically determines whether the equipment requires maintenance, along with the corresponding maintenance schedule and strategy. After each maintenance operation, the equipment's maintenance strategy reaches a new probabilistic state. Traditional prediction algorithms, due to manual feature extraction, have a degree of uncertainty, making them difficult to generalize. This type of time series-based decision-making problem is well-suited for modeling using the LSTM algorithm. This modeling eliminates the reliance on manual signal feature extraction, addressing the diversity, complexity, and high modeling costs of aerospace equipment.

[0128] The algorithm primarily uses an LSTM neural network. Sensor data from each device in each time window serves as classifier input, while failure probability and lifespan predictions serve as classifier output, resulting in corresponding maintenance decisions. The dynamic maintenance decision framework for failure and lifespan prediction based on real-time data is shown in the figure below.

[0129] Historical data from equipment sensors, fault status, and maintenance decision history serve as input for classifier training. The data is first normalized and mapped to the [0, 1] range. The three types of data are then aligned based on the equipment ID and time labels and fed into the LSTM classifier for training. The dataset is divided into training, validation, and test sets in an 8:1:1 ratio. The weights for the input, forget, and output gates of the LSTM classifier are learned. The LSTM classifier consists of an input layer, a hidden layer, and an output layer. The hidden layer consists of two layers of 50 neurons, and the optimization objective is to minimize the mean squared error. The output layer uses a sigmoid activation function to output a failure probability lifetime prediction and the corresponding maintenance decision.

[0130] The above two key technologies provide technical guarantees for the innovative points of the intelligent predictive maintenance system for aerospace equipment, and also systematically realize the intelligent predictive maintenance of aerospace equipment, realizing the development and innovation of aerospace equipment maintenance.

[0131] The advantages of the present invention are:

[0132] 1) The integrated platform design covers the entire process of aerospace equipment repair and maintenance, including monitoring, processing analysis, fault prediction and diagnosis, health life prediction, result visualization, and support decision-making and management optimization, providing great convenience for aerospace equipment maintenance and management.

[0133] 2) Data-driven predictive maintenance modeling has certain versatility and is well adaptable to the diversity and complexity of aerospace equipment.

[0134] 3) The system design modules are clearly divided, which facilitates the addition, access and modification of equipment, and has high maintainability and scalability.

[0135] 4) The visualization module provides intuitive result analysis and display functions, providing intuitive result display for professionals or non-professionals to facilitate user decision-making.

[0136] 5) Based on advanced Internet of Things, data processing, and artificial intelligence technologies, the accuracy of predictive maintenance of aerospace equipment has been effectively improved, the ability to identify faults has been greatly enhanced, and more reasonable decision-making recommendations have been provided.

[0137] In summary, the present invention provides a one-stop predictive maintenance system for aerospace equipment, which solves the problems of delayed planned maintenance of aerospace equipment affecting launch missions, the lack of universality of general reliability modeling maintenance, and the lack of intelligent guarantee decision-making and management optimization process in the current equipment maintenance system, making aerospace equipment maintenance more intelligent and complete.

[0138] Although the present invention has been disclosed above in terms of preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art may make possible changes and modifications to the technical solutions of the present invention by using the methods and technical contents disclosed above without departing from the spirit and scope of the present invention. Therefore, any simple modifications, equivalent changes and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the technical solutions of the present invention shall fall within the scope of protection of the technical solutions of the present invention.

Claims

1. A predictive maintenance system for aerospace equipment, characterized by It includes perception subsystem, processing and analysis subsystem, fault prediction and diagnosis subsystem, health and life prediction subsystem, visualization subsystem, and security decision-making management optimization subsystem; The sensing system collects the operating parameters of aerospace equipment and transmits them to the processing and analysis subsystem; The processing and analysis subsystem processes the operating parameters of the aerospace equipment to obtain characteristic data reflecting the status of the aerospace equipment, stores the characteristic data reflecting the status of the aerospace equipment in the parameter database, and sends it to the fault prediction and diagnosis subsystem and the health and life prediction subsystem at the same time; The fault prediction and diagnosis subsystem performs fault diagnosis and analysis on characteristic data reflecting the status of aerospace equipment to obtain the most likely type of fault and its probability of occurrence at the current moment; predicts the fault trend of aerospace equipment in a preset time period in the future to obtain the type of aerospace equipment fault and its probability of occurrence in the preset time period in the future, and simultaneously sends the type of aerospace equipment fault and its probability of occurrence at the current moment and in the preset time period in the future to the health and life prediction subsystem; The health and life prediction subsystem evaluates and predicts the health and life of aerospace equipment based on characteristic data reflecting the equipment's status, the types of aerospace equipment failures at the current moment and within a preset time period in the future, and their probability of occurrence, thereby obtaining a health rating and life curve for the equipment. The visualization system displays the process data and processing results of the processing and analysis subsystem, fault prediction and diagnosis subsystem, and health and life prediction subsystem in a visual way; Ensure the decision-making management optimization subsystem and obtain the processing results of the processing and analysis subsystem, that is, the characteristic data C reflecting the status of aerospace equipment in the past preset time period t , the processing result of the fault prediction and diagnosis subsystem, that is, the fault probability F within the future preset time period t , the processing results of the health and life prediction subsystem, that is, the health rating L in the future preset time period t , forming a multi-source data matrix, multiplying the multi-source data matrix by the input weight matrix for weighting, and using the weighted multi-source data matrix as the input vector The multi-input multi-output LSTM neural network is used to predict the time m when aerospace equipment failure may occur. f , the most cost-effective maintenance time opt , the best maintenance and repair method a, and plan resource management based on the time when aerospace equipment failures may occur, and obtain a form of the number of aerospace equipment spare parts required within a preset time period; The multi-input multi-output LSTM neural network includes an input layer, a hidden layer, and an output layer, wherein the input layer tensor is a three-dimensional matrix consisting of the number of samples, the time step, and the sample dimension. for: Among them, w c is the state feature sequence weight, w f is the fault data weight, w l is the weight of healthy life expectancy data; samples is the number of samples; There are two hidden layers, denoted as the first hidden layer and the second hidden layer. Each layer consists of 50 neurons. Each neuron contains an input gate, a forget gate, and an output gate. The input gate, forget gate, and output gate are set with different weights. The activation function of the output layer adopts the sigmoid function, and the minimization of the mean square error is used as the optimization goal. The result of the hidden layer is mapped to the output and the maintenance decision data H is obtained. t , the maintenance decision data H t for: in, The time when aerospace equipment failure may occur, For the most cost-effective maintenance time, a t As a means of maintenance.

2. The predictive maintenance system for aerospace equipment according to claim 1, characterized in that: For mechanical structure aerospace equipment, the state feature sequence weight w c is 0.6, and the healthy life expectancy data w l weight and fault data weight w f 0.2, 0.2 respectively; For hydraulic system equipment, the health life data weight w l is 0.6, and the state feature sequence weight w c and fault data weight w f 0.2, 0.2 respectively; For control system equipment, the fault data weight w f is 0.6, and the state feature sequence weight w c and healthy life expectancy data w l They are 0.2 and 0.2 respectively.

3. The aerospace equipment predictive maintenance system according to claim 2, characterized in that The multi-input multi-output LSTM neural network is trained by the following method: S1. Based on the equipment number tag and time tag, obtain the historical data C reflecting the status characteristics of aerospace equipment t ′, Failure probability historical data F t ′, health rating historical data L t ′, and its corresponding time m when aerospace equipment failure may occur f ′、The most cost-effective maintenance time m opt ', optimal maintenance and repair method a'; S2, will reflect the historical data of aerospace equipment status characteristics C t ′, Failure probability historical data F t ′, health rating historical data L tj Perform input weighting processing respectively to obtain the input vector And the input vector Perform normalization processing to map it to the [0,1] interval; S3. Change the time period corresponding to the historical data, repeat steps S1 and S2 to obtain the data set. J is the sample size; S4. Dataset The training set, validation set and test set are divided into 8:1:1 ratios, and the multi-input multi-output LSTM neural network is trained to determine the input gate weight, forget gate weight and output gate weight of each neuron in the hidden layer of the multi-input multi-output LSTM neural network, thereby determining the multi-input multi-output LSTM neural network.

4. The aerospace equipment predictive maintenance system according to claim 1, characterized in that Health and lifespan prediction subsystem, including health status prediction model and data-driven RUL model module; The health status prediction model module uses a linear regression model, a logistic regression model, or a Gaussian process regression model to fit the health status of aerospace equipment based on historical characteristic data reflecting the status of aerospace equipment. This module obtains a curve showing the health status of aerospace equipment over time and completes the assessment and prediction of the health status of aerospace equipment. The data-driven RUL model module uses a Bayesian estimation model to evaluate and predict the remaining service life based on the historical life data of the equipment and the health status change curve over time, obtains the remaining service life curve, and sends the remaining service life curve to the equipment life prediction model module.

5. The aerospace equipment predictive maintenance system according to claim 1, characterized in that The health and life prediction subsystem also includes a performance failure degradation model module; The performance failure degradation model is based on the historical characteristic data reflecting the status of aerospace equipment. It uses a linear degradation model or a power degradation model to fit the equipment degradation process, obtain the equipment degradation curve, and predict the equipment degradation trend.

6. The aerospace equipment predictive maintenance system according to claim 1, characterized in that The aerospace equipment operating parameters include equipment performance data, equipment input data, equipment output data, equipment structural data, equipment environment data, and equipment alarm data.

Images