An artificial intelligence-based electromechanical system health monitoring and optimization method

Abstract

The application belongs to the technical field of intelligent monitoring and operation and maintenance management of electromechanical equipment, and discloses a health monitoring and optimization method for electromechanical systems based on artificial intelligence. The health monitoring and optimization method for electromechanical systems based on artificial intelligence builds a digital twin architecture based on a simulation model, reasonably arranges multi-source sensors, and comprehensively covers key equipment. The system collects operation data at a high frequency, synchronously transmits the operation data to the cloud through multiple channels, realizes real-time monitoring, collects data, pre-processes the data, removes abnormalities, and ensures data quality. The cleaned data is input into a neural network, features are extracted, operation trends and failure modes are identified, a prediction model is established in combination with historical data, a maintenance knowledge base is constructed, the state of equipment is predicted in real time, early warnings are dynamically generated, the health and carbon emission states are mapped in real time through virtual-real synchronization, the operation of equipment is intelligently optimized, and the cloud closed-loop monitoring is continuously iterated.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of intelligent monitoring and operation and maintenance management of electromechanical equipment, and particularly relates to an electromechanical system health monitoring and optimization method based on artificial intelligence. BACKGROUND

[0002] With the deep penetration of smart city construction and green low-carbon concept, the operation and management of infrastructure electromechanical systems are in a complex predicament. The traditional operation and maintenance mode takes structural safety as the core concern, but pays insufficient attention to real-time monitoring and preventive maintenance of the health status of electromechanical systems. This "emphasis on structure and neglect of electromechanical" thinking makes equipment frequently malfunction during operation, and the operation and maintenance team often responds passively after the fault occurs, which not only increases maintenance costs, but also may affect the normal operation of infrastructure.

[0003] The existing monitoring system also has obvious shortcomings. It mainly collects single parameters and cannot comprehensively perceive multi-dimensional states such as equipment vibration, temperature, electrical parameters and fluid parameters. There is a lack of effective linkage between monitoring modules, data forms an information island, resulting in a significant reduction in monitoring accuracy and difficulty in accurately predicting potential equipment failures. In addition, current carbon emission management only focuses on the operation stage of electromechanical systems, ignoring the carbon emission situation of equipment throughout its life cycle from production, transportation, installation to scrap. Under this management mode, it is impossible to dynamically track and optimize control of carbon emissions throughout the life cycle of equipment. SUMMARY

[0004] The purpose of the present application is to provide an electromechanical system health monitoring and optimization method based on artificial intelligence to solve the problems raised in the background.

[0005] In order to achieve the above purpose, the present application provides the following technical solution: an electromechanical system health monitoring and optimization method based on artificial intelligence, which builds a digital twin architecture based on a simulation model, reasonably arranges multi-source sensors, and comprehensively covers key equipment. The system collects high-frequency operation data, which is transmitted to the cloud through multi-channel synchronization to realize real-time monitoring. The collected data is preprocessed to remove anomalies and ensure data quality. The cleaned data is input into a neural network to extract features, identify operation trends and failure modes. A prediction model is established based on historical data to build a maintenance knowledge base. The device state is predicted in real time, and early warning is dynamically generated. Through virtual-real synchronization, the health and carbon emission states are mapped in real time, the device operation is intelligently optimized, and the cloud closed-loop monitoring is continuously iterated. The specific steps of the optimization method are as follows:

[0006] S1: Multi-source sensor network construction and equipment monitoring object determination: Based on the finite element analysis model, transient simulation model and carbon emission calculation model of building and mechanical and electrical system, the digital twin infrastructure of mechanical and electrical system is established. Through model simulation and working condition analysis, key monitoring objects and representative parts prone to failure are selected to determine monitoring points. For different equipment and environmental parameters, NB-IoT wireless vibration sensors (sampling frequency not less than 500 Hz), thermal imaging sensors (resolution not less than 160x120 pixels, refresh rate ≥9 Hz), current and voltage sensors (data refresh period ≤1 s), flow and pressure sensors (response time ≤0.5 s) and environmental sensors are arranged. The number of measuring points is not less than 3 groups of sensor units per 100㎡ of building area to ensure comprehensive data coverage;

[0007] S2: Real-time data high-frequency acquisition and multi-channel wireless transmission: The collected multi-source monitoring data is transmitted through NB-IoT (delay ≤2 s), 5G (delay ≤0.5 s) and RS485 bus (data transmission rate ≥115.2 kbps) in asynchronous multi-channel real-time transmission. Based on the transmission architecture combining wireless and wired transmission, vibration, temperature, current, voltage, flow, pressure, humidity, illumination and other multi-parameters during the operation of mechanical and electrical system are collected in high frequency and real time, and multi-channel data is transmitted to the cloud or edge computing platform simultaneously to realize data continuity, stability and high timeliness;

[0008] S3: Health monitoring raw data cleaning and abnormal value preprocessing: For the real-time collected multi-dimensional data, sliding window analysis, time domain average, wavelet denoising and other data processing techniques are used to complete abnormal value elimination, data completion and standardization processing to ensure the integrity, accuracy and time sequence continuity of health monitoring data, and eliminate the interference of environmental noise and load fluctuation on subsequent analysis;

[0009] S4: Mechanical and electrical system operation state feature extraction and multi-scale trend analysis: Three-layer convolutional neural network (convolution kernel size 3x3, step 1, output channel number 32, 64, 128) is used to extract deep features of vibration, heat, electricity and fluid parameters. The peak value, root mean square, kurtosis and spectral main peak displacement of time domain vibration signal are extracted, the total harmonic distortion (THD) and current impact characteristics of electrical parameters are extracted, the maximum temperature difference of high temperature area of thermal imaging is extracted, and the flow fluctuation rate of fluid parameters is extracted. The device operation trend is calculated through time series data flow to identify potential failure modes;

[0010] S5: Historical state modeling and predictive maintenance knowledge base construction: Based on the collected historical equipment state data, an electromechanical equipment state monitoring knowledge base is established, and a data-driven support vector machine (SVM) and long short-term memory network (LSTM) prediction model is constructed, the input features include: vibration root mean square, harmonic distortion rate, thermal imaging high temperature point temperature, flow fluctuation rate. The amount of training set data is ≥100,000 groups, the prediction error target is controlled within 5%, and the output is the remaining service life and health index of the equipment. Simultaneously, a device fault case knowledge base and operation and maintenance response strategy library are constructed;

[0011] S6: Real-time state prediction and hierarchical dynamic early warning mechanism: The LSTM model is used to predict the trend of the equipment running state in the next 24 hours, and typical fault signs such as sudden increase in vibration, current distortion, and abnormal temperature rise in thermal imaging are monitored. Set three dynamic early warning thresholds: mild abnormality (prediction deviation 5%-10%), moderate abnormality (prediction deviation 10%-15%), and severe abnormality (prediction deviation ≥15%), the early warning generation time lag is controlled to be ≤1 second, and the system supports alarm linkage and on-site inspection dispatching functions;

[0012] S7: Digital twin virtual-real synchronization and full life cycle carbon emission mapping: Through the bidirectional fusion of finite element simulation model and real-time monitoring data, a digital twin platform of electromechanical system is established, ensuring that the virtual-real data synchronization delay is ≤2 seconds, and the virtual-real residual error is ≤5%. Based on the real-time load, energy efficiency coefficient and running time of the equipment, the carbon emission of a single equipment is dynamically calculated, and the full life cycle carbon emission curve of the electromechanical system is accumulated. Update the device health index and carbon emission prediction results synchronously;

[0013] S8: Intelligent optimization scheduling and priority operation strategy implementation: Based on the device health index, energy efficiency parameter and carbon emission index of the digital twin platform, dynamically adjust the start-stop sequence, load distribution and operation priority of the equipment, preferentially schedule the equipment with good health status and higher energy efficiency, realize the optimization of system overall operation efficiency and minimization of carbon emission, and avoid the continuous fatigue of high-load running equipment;

[0014] S9: Cloud remote operation and maintenance monitoring and closed-loop feedback optimization: Through the Web, APP or local control terminal, real-time display of equipment health status, warning information, carbon emission and scheduling optimization results. Synchronize cloud operation and maintenance decision and on-site control system, form intelligent operation and maintenance closed-loop management, continuously iterate and optimize prediction model and scheduling strategy combined with new equipment operation data, and improve the full life cycle health management capability of electromechanical system.

[0015] Preferably, the specific steps of the multi-source sensor network construction and equipment monitoring object determination in S1 are as follows:

[0016] Step one: Digital twin architecture building and monitoring planning: Based on the finite element analysis, transient simulation and carbon emission calculation model of building and mechanical and electrical system, the digital twin basic architecture of mechanical and electrical system is built. Through deep simulation and working condition analysis of the model, key monitoring objects and vulnerable parts are accurately selected, and monitoring points are scientifically determined, providing reliable basis for real-time perception and accurate control of mechanical and electrical system operation state, and laying a solid foundation for digital twin application;

[0017] Step two: Multi-source sensor layout and measurement point optimization: According to the differences of equipment characteristics and environmental parameters, reasonably deploy NB-IoT wireless vibration, thermal imaging, current and voltage and other multi-type sensors, and clearly define the performance parameter requirements of each sensor. At the same time, follow the principle of configuring not less than 3 groups of sensor units per 100㎡ of building area, and dynamically plan the number of measurement points considering factors such as building area, to comprehensively guarantee the integrity and effectiveness of monitoring data coverage;

[0018] The expression formula of multi-source sensor measurement point optimization is:

[0019] ;

[0020] In the formula, Minimum number of sensors (units) Monitoring area (㎡) Ceiling function;

[0021] Preferably, the specific steps of real-time data high-frequency collection and multi-channel wireless transmission in S2 are as follows:

[0022] Step one: Multi-channel hybrid transmission architecture building: Adopting asynchronous multi-channel transmission mode combining NB-IoT, 5G and RS485 bus, build a wireless and wired complementary data transmission architecture. NB-IoT ensures wireless transmission efficiency with a delay of no more than 2s, 5G ensures low delay with a delay of ≤0.5s, and RS485 bus ensures stable wired transmission with a data transmission rate of ≥115.2kbps. The three transmission modes work together to build an efficient and stable channel foundation for mechanical and electrical system data transmission;

[0023] Step two: Multi-parameter high-frequency collection and data synchronization: Based on the above transmission architecture, the vibration and temperature multi-parameters in the operation of mechanical and electrical system are collected in high frequency and real time. Using multi-channel data synchronization logic, the collected data of various types are quickly and stably transmitted to the cloud or edge computing platform through different transmission channels. This logic fully considers the characteristics of each transmission channel, ensuring the high frequency of data collection and the continuity, stability and high timeliness of transmission, providing reliable data support for system analysis and decision-making;

[0024] The expression formula of multi-channel data synchronization is:

[0025] ;

[0026] In the formula, Time Synchronization data set of time, The Channel sensor delay synchronized data, The Channel transmission delay (seconds), Total number of sensor channels, multi-channel synchronization data needs to be compensated by delay before being time-stamped.

[0027] Preferably, the specific steps of S3 health monitoring raw data cleaning and outlier preprocessing are as follows:

[0028] Step one: Application of multi-dimensional data processing technology: for real-time collected multi-dimensional data of mechanical and electrical system, comprehensive use of sliding window analysis, time domain average, wavelet denoising data processing technology, through sliding window analysis to dynamically capture data characteristics, use time domain average to integrate data information, and use wavelet denoising to filter interference components in the data. In this process, according to the abnormal data elimination logic, the abnormal values are accurately identified and eliminated, and the data completion and standardization processing are carried out, and the original data is comprehensively optimized;

[0029] Step two: Data quality assurance and interference elimination: through a series of data processing operations, the integrity, accuracy and time sequence continuity of health monitoring data are effectively ensured, the application of abnormal data elimination logic avoids the influence of abnormal values on the whole data; data completion and standardization processing perfect the data content and unify the format, which eliminates the interference of environmental noise, load fluctuation and other factors on subsequent data analysis, and provides reliable data basis for mechanical and electrical system state evaluation, fault diagnosis and other data-based analysis;

[0030] The expression formula of abnormal data elimination is:

[0031] ;

[0032] In the formula, Standard deviation score (Z-score), Collected data, Sliding window mean, Sliding window standard deviation, abnormal data judgment: when ∣Z∣>3, the data is judged as abnormal and needs to be eliminated.

[0033] Preferably, the specific steps of S4 mechanical and electrical system running state feature extraction and multi-scale trend analysis are as follows:

[0034] Step 1: Multi-parameter deep feature extraction: Using a three-layer convolutional neural network with 3×3 kernels, a stride of 1, and output channels of 32, 64, and 128 respectively, deep feature mining is performed on four types of parameters: vibration, heat, electricity, and fluid. For time-domain vibration signals, peak value, root mean square, and other features are extracted according to specific calculation logic; for electrical parameters, the total harmonic distortion rate and current impact are focused; for thermal imaging, the maximum temperature difference in high-temperature areas is considered; and for fluid parameters, flow fluctuation rate is extracted, thus achieving feature integration of multi-source data.

[0035] Step 2: Equipment Failure Mode Identification: The extracted multi-parameter features are converted into time-series data streams. Based on the data stream analysis, the equipment operation trend is analyzed. Through continuous monitoring and analysis of various parameter features, combined with the changes in key indicators of the root mean square of vibration signals, potential equipment failure modes are accurately identified, and the equipment operating status is predicted in advance, providing strong support for electromechanical system maintenance and fault early warning.

[0036] The formula for the root mean square of the vibration signal is:

[0037] ;

[0038] In the formula, RMS: root mean square value of the vibration signal. No. Vibration acceleration at each sampling point The total number of sampling points and the root mean square of vibration are key indicators of equipment health status, used to monitor equipment vibration levels in real time.

[0039] Preferably, the specific steps for constructing the historical state modeling and predictive maintenance knowledge base in S5 are as follows:

[0040] Step 1: Predictive Model Construction and Training: Based on the collected historical equipment status data, a knowledge base for electromechanical equipment status monitoring is built. Support Vector Machine and Long Short-Term Memory Network (LSTM) are used as inputs to construct a data-driven prediction model. The model is deeply trained using no less than 100,000 sets of training data. The LSTM status prediction logic is used to optimize the model, strictly controlling the prediction error to within 5%, and accurately outputting the remaining service life and health index of the equipment.

[0041] Step 2: Collaborative Construction of Knowledge Base System: While completing the construction of the prediction model, a knowledge base for equipment failure cases and a maintenance response strategy base are built simultaneously. The failure case knowledge base summarizes historical failure information to provide a reference for problem diagnosis; the maintenance response strategy base formulates targeted handling solutions to achieve full-process knowledge collaboration from failure prediction and diagnosis to handling, and improve the intelligent maintenance system for electromechanical equipment.

[0042] The formula for LSTM state prediction is:

[0043] ;

[0044] wherein, current time hidden state (predicted output), previous time hidden state, current time input features (including vibration, electrical parameters, thermal imaging, etc.), weight matrix, bias term; single-step prediction structure based on long short-term memory network, applied to future state prediction.

[0045] Preferably, the specific steps of real-time state prediction and hierarchical dynamic early warning mechanism in S6 are as follows:

[0046] Step one: LSTM model trend prediction and fault monitoring: use LSTM model to carry out trend prediction on the future 24-hour operation state of the equipment, focus on capturing typical fault signs such as vibration surge, current distortion, and abnormal temperature rise in thermal imaging, process time series data through the state prediction logic of LSTM, realize accurate deduction of the operation trend of the equipment, and provide reliable prediction basis for fault warning;

[0047] Step two: three-level dynamic early warning mechanism and system function: set a three-level dynamic early warning threshold system: mild abnormality (prediction deviation 5%~10%), moderate abnormality (10%~15%), and serious abnormality (≥15%), control the early warning generation lag time within 1 second, this mechanism combines dynamic threshold logic, automatically triggers the corresponding warning level according to the prediction deviation, and at the same time, the system supports alarm linkage and on-site inspection dispatching functions, forming a closed-loop management from prediction to disposal;

[0048] The expression formula of the three-level dynamic early warning threshold is:

[0049] ;

[0050] wherein, prediction error percentage, predicted state value (such as predicted vibration), current actual monitoring state value;

[0051] Warning level: mild abnormality: ; moderate abnormality: ; serious abnormality: , based on the prediction error, dynamically divide the warning level, and timely prompt the change of equipment state.

[0052] Preferably, the specific steps of digital twin virtual-real synchronization and full life cycle carbon emission mapping in S7 are as follows:

[0053] Step one: Building a digital twin platform for the electromechanical system: Through the bidirectional fusion of finite element simulation models and real-time monitoring data, a digital twin platform for the electromechanical system is constructed. Relying on the virtual-real residual error control logic, the data synchronization quality is strictly controlled to ensure that the virtual-real data synchronization delay is ≤2 seconds and the virtual-real residual error is ≤5%. This achieves precise mapping and dynamic linkage between the physical system and the virtual model.

[0054] Step two: Dynamic management of carbon emissions and health index: Based on real-time load, energy efficiency coefficient, and running time of the equipment, the carbon emissions of a single device are dynamically calculated, and the full life cycle carbon emission curve of the electromechanical system is accumulated. At the same time, the platform updates the equipment health index and carbon emission prediction results in real time, achieving collaborative monitoring and management of equipment operating status and environmental protection indicators.

[0055] The digital twin virtual-real residual error control expression is:

[0056] ;

[0057] In the formula, Virtual-real residual error percentage, Digital twin virtual state data, Real-time actual monitoring data.

[0058] Preferably, the specific steps of the intelligent optimization scheduling and priority operation strategy implementation in S8 are as follows:

[0059] Step one: Developing a dynamic scheduling strategy: Based on the equipment health index, energy efficiency parameters, and carbon emission indicators of the digital twin platform, a dynamic scheduling scheme is developed. According to the health priority scheduling priority logic, the device start-stop sequence, load distribution, and operation priority are determined, and the devices with good health status and high energy efficiency are prioritized for scheduling, laying the foundation for efficient operation of the system.

[0060] Step two: Achieving system operation optimization goals: Through the above dynamic scheduling, the overall system operation efficiency is optimized and carbon emissions are minimized. Reasonable load distribution avoids continuous fatigue of high-load devices, prolonging the service life of the equipment. At the same time, under the premise of ensuring stable operation of the system, the efficiency and environmental protection are balanced, achieving comprehensive benefit improvement.

[0061] The health priority scheduling priority expression formula is:

[0062] ;

[0063] In the formula, Device operation priority score, Device health index (0-1), Current energy efficiency coefficient of the device (0-1), Current carbon emissions per unit time of the device (g / kWh), ).

[0064] Preferably, the specific steps of cloud remote operation and maintenance monitoring and closed-loop feedback optimization in S9 are as follows:

[0065] Step one: multi-terminal information display and operation closed loop: through the Web terminal, APP terminal and local control terminal, the device health status, early warning information, carbon emission data and scheduling optimization results are presented in real time, the cloud operation decision and the field control system are synchronized, the time control logic of cloud alarm response is followed, the intelligent operation and maintenance closed loop management is formed, and the efficient connection of operation and maintenance is ensured;

[0066] Step two: model iteration and management ability improvement: combining new equipment operation data, continuously iterating and optimizing the prediction model and scheduling strategy, through continuously improving the model precision and strategy adaptability, gradually improving the health management ability of the mechanical and electrical system in the whole life cycle, realizing the whole process efficient management from real-time monitoring, intelligent decision to continuous optimization;

[0067] The cloud alarm response delay expression formula is:

[0068] ;

[0069] In the formula, Total alarm response delay (seconds), Data upload transmission delay (seconds), Cloud processing calculation delay (seconds), Cloud to client alarm push delay (seconds), ensure Ttotal≤3 seconds, meet the requirements of cloud real-time alarm response.

[0070] The beneficial effects of the present application are as follows:

[0071] 1、The present application can realize real-time collection and deep learning analysis of key data such as vibration, electrical parameter and thermal imaging of mechanical and electrical system in actual operation by constructing multi-dimensional state feature extraction and predictive maintenance algorithm based on artificial intelligence, extract equipment operation trend and potential fault features through the joint application of convolutional neural network and long short-term memory network, realize accurate prediction of equipment state, compared with the traditional mode of relying on periodic maintenance or post-maintenance in the prior art, the present application can significantly identify equipment abnormalities in advance, effectively reduce the probability of sudden failure, reduce the demand for emergency repair, thereby improving the continuity and stability of system operation, prolonging the service life of equipment, and optimizing the operation and maintenance efficiency of mechanical and electrical system.

[0072] 2、The present application breaks through the single data acquisition mode of the existing monitoring system by adopting multi-source sensor fusion technology, establishes a comprehensive health state perception system through synchronous layout of vibration sensors, thermal imaging sensors, current and voltage monitoring units, flow and pressure sensors and environmental parameter monitoring equipment, and realizes high-frequency real-time acquisition of multi-channel data and cloud-edge collaborative processing by fusing NB-IoT, 5G and wired networks, solving the problems of scattered traditional energy consumption monitoring and health monitoring data, lagging response and untimely cleaning; through sliding window, wavelet denoising, dynamic abnormality elimination and other cleaning technologies, the real-time, integrity and high-precision input of data are guaranteed, and the overall monitoring response speed and fault diagnosis capability of the system are effectively improved.

[0073] 3、The present application breaks through the single data acquisition mode of the existing monitoring system by adopting multi-source sensor fusion technology, establishes a comprehensive health state perception system through synchronous layout of vibration sensors, thermal imaging sensors, current and voltage monitoring units, flow and pressure sensors and environmental parameter monitoring equipment, and realizes high-frequency real-time acquisition of multi-channel data and cloud-edge collaborative processing by fusing NB-IoT, 5G and wired networks, solving the problems of scattered traditional energy consumption monitoring and health monitoring data, lagging response and untimely cleaning; through sliding window, wavelet denoising, dynamic abnormality elimination and other cleaning technologies, the real-time, integrity and high-precision input of data are guaranteed, and the overall monitoring response speed and fault diagnosis capability of the system are effectively improved. BRIEF DESCRIPTION OF DRAWINGS

[0074] Figure 1 The present application is based on artificial intelligence-based mechanical and electrical system health monitoring and optimization method flowchart. DETAILED DESCRIPTION

[0075] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0076] As shown in Figure 1 The present application provides an artificial intelligence-based mechanical and electrical system health monitoring and optimization method, and the specific steps of the artificial intelligence-based mechanical and electrical system health monitoring and optimization method are as follows:

[0077] S1: Multi-source sensor network construction and equipment monitoring object determination: Based on the finite element analysis model, transient simulation model and carbon emission calculation model of building and mechanical and electrical system, the digital twin infrastructure of mechanical and electrical system is established. Through model simulation and working condition analysis, key monitoring objects and representative parts prone to failure are selected to determine monitoring points. For different equipment and environmental parameters, NB-IoT wireless vibration sensors (sampling frequency not less than 500 Hz), thermal imaging sensors (resolution not less than 160x120 pixels, refresh rate ≥9 Hz), current and voltage sensors (data refresh period ≤1 s), flow and pressure sensors (response time ≤0.5 s) and environmental sensors are arranged. The number of measuring points is not less than 3 groups of sensor units per 100㎡ of building area to ensure comprehensive data coverage;

[0078] S2: Real-time data high-frequency acquisition and multi-channel wireless transmission: The collected multi-source monitoring data is transmitted through NB-IoT (delay ≤2s), 5G (delay ≤0.5s) and RS485 bus (data transmission rate ≥115.2kbps) in asynchronous multi-channel real-time transmission. Based on the transmission architecture combining wireless and wired transmission, vibration, temperature, current, voltage, flow, pressure, humidity, illumination and other multi-parameters during the operation of mechanical and electrical system are collected in high frequency and real time, and multi-channel data is transmitted to the cloud or edge computing platform simultaneously to realize data continuity, stability and high timeliness;

[0079] S3: Health monitoring raw data cleaning and abnormal value preprocessing: For real-time collected multi-dimensional data, sliding window analysis, time domain average, wavelet denoising and other data processing techniques are used to complete abnormal value elimination, data completion and standardization processing to ensure the integrity, accuracy and time sequence continuity of health monitoring data, and eliminate the interference of environmental noise and load fluctuation on subsequent analysis;

[0080] S4: Mechanical and electrical system operation state feature extraction and multi-scale trend analysis: Three-layer convolutional neural network (convolution kernel size 3x3, step 1, output channel number 32, 64, 128) is used to extract deep features of vibration, heat, electricity and fluid parameters. The peak value, root mean square, kurtosis, frequency spectrum main peak displacement of time domain vibration signal, the total harmonic distortion (THD) and current impact characteristics of electrical parameters, the maximum temperature difference of high temperature area of thermal imaging, and the flow fluctuation rate of fluid parameters are extracted. Through time series data flow calculation, the equipment operation trend is identified to recognize potential failure modes;

[0081] S5: Historical state modeling and predictive maintenance knowledge base construction: Based on the collected historical equipment state data, a mechanical and electrical equipment state monitoring knowledge base is established, a data-driven support vector machine (SVM) and long short-term memory network (LSTM) prediction model is constructed, and the input features include: vibration root mean square, harmonic distortion rate, thermal imaging high temperature point temperature, and flow fluctuation rate. The training set data amount is greater than or equal to 100,000 groups, the prediction error target is controlled within 5%, and the output is the remaining service life and health index of the equipment. A device fault case knowledge base and operation and maintenance response strategy library are constructed at the same time;

[0082] S6: Real-time state prediction and hierarchical dynamic early warning mechanism: The LSTM model is used to predict the trend of the equipment running state in the next 24 hours, and typical fault signs such as sudden increase in vibration, current distortion, and abnormal temperature rise in thermal imaging are monitored. Three levels of dynamic early warning thresholds are set: mild abnormality (prediction deviation 5%-10%), moderate abnormality (prediction deviation 10%-15%), and severe abnormality (prediction deviation ≥15%), the early warning generation time lag is controlled to be ≤1 second, and the system supports alarm linkage and on-site inspection dispatching functions;

[0083] S7: Digital twin virtual-real synchronization and full life cycle carbon emission mapping: Through the bidirectional fusion of finite element simulation model and real-time monitoring data, a mechanical and electrical system digital twin platform is established, ensuring that the virtual-real data synchronization delay is ≤2 seconds and the virtual-real residual error is ≤5%. Based on the real-time load, energy efficiency coefficient and running time of the equipment, the carbon emission of a single equipment is dynamically calculated, and the full life cycle carbon emission curve of the mechanical and electrical system is accumulated. The health index of the equipment and the carbon emission prediction results are updated synchronously;

[0084] S8: Intelligent optimization scheduling and priority operation strategy implementation: Based on the health index, energy efficiency parameter and carbon emission index of the digital twin platform, the device start-stop sequence, load distribution and operation priority are dynamically adjusted, the devices with good health status and higher energy efficiency are preferentially scheduled, the overall operation efficiency of the system is optimized, and the carbon emission is minimized, and the continuous fatigue of high-load running equipment is avoided;

[0085] S9: Cloud remote operation and maintenance monitoring and closed-loop feedback optimization: Through the Web, APP or local control terminal, the health status of the equipment, early warning information, carbon emission and scheduling optimization results are displayed in real time. The cloud operation and maintenance decision and the on-site control system are synchronized to form an intelligent operation and maintenance closed-loop management, and the prediction model and scheduling strategy are continuously iterated and optimized based on the new data of the subsequent equipment operation, and the full life cycle health management capability of the mechanical and electrical system is improved;

[0086] The application also continuously collects real-time operation data of the equipment through cloud remote monitoring and intelligent closed-loop feedback mechanism, supports dynamic iteration and self-optimization of the prediction model and scheduling strategy, further improves the intelligentization and self-adaptation capability of the system, and is verified through multiple scene applications;

[0087] It can adapt to complex environment, multi-type equipment operation demand, has good engineering implementability and expansibility, the system not only can significantly reduce the whole life cycle operation and maintenance cost of mechanical and electrical equipment, but also can improve the digitalization and intelligent level of equipment management, provides efficient, low carbon, safe intelligent operation and maintenance solution for important infrastructure such as intelligent building, rail transit, airport hub, green park, has good popularization prospect and social and economic value.

[0088] Among them, the multi-source sensor network construction in S1 and the determination of the equipment monitoring object means relying on the finite element analysis, transient simulation and carbon emission calculation model of building and mechanical and electrical system, constructing the digital twin basic framework of mechanical and electrical system, through the deep simulation and working condition analysis of the model, accurately screening the key monitoring objects and the easy fault parts, scientifically determining the monitoring points, providing reliable basis for the real-time perception and accurate control of the subsequent mechanical and electrical system operation state, building the application foundation of digital twin;

[0089] According to the difference of equipment characteristics and environmental parameters, reasonably deploy NB-IoT wireless vibration, thermal imaging, current and voltage and other multi-type sensors, and clearly define the performance parameter requirements of each sensor. At the same time, follow the principle of configuring not less than 3 groups of sensor units per 100㎡ building area, combined with multi-source sensor measurement point optimization logic, comprehensively consider factors such as building area to dynamically plan the number of measurement points, and comprehensively guarantee the integrity and effectiveness of monitoring data coverage;

[0090] Among them, the real-time data high-frequency collection and multi-channel wireless transmission in S2 means adopting asynchronous multi-channel transmission mode combining NB-IoT, 5G and RS485 bus, building a wireless and wired complementary data transmission architecture. NB-IoT guarantees wireless transmission efficiency with a delay of not more than 2s, 5G guarantees low delay with a delay of ≤0.5s, and RS485 bus ensures stable wired transmission with a data transmission rate of ≥115.2kbps. The three transmission modes work together to build an efficient and stable channel foundation for mechanical and electrical system data transmission;

[0091] Based on the above transmission architecture, the vibration, temperature and other parameters in the operation of mechanical and electrical system are collected in real time with high frequency. The multi-channel data synchronization logic is used to quickly and stably synchronize the collected data of various types to the cloud or edge computing platform through different transmission channels. This logic fully considers the characteristics of each transmission channel to ensure the high frequency of data collection and the continuity, stability and high timeliness of transmission, providing reliable data support for system analysis and decision-making;

[0092] The health monitoring original data cleaning and abnormal value preprocessing in S3 refers to the comprehensive use of sliding window analysis, time domain average and wavelet denoising data processing technology for real-time collected mechanical and electrical system multi-dimensional data. The data characteristics are dynamically captured by sliding window analysis, the data information is integrated by time domain average, and the interference components in the data are filtered by wavelet denoising. In this process, the abnormal values are accurately identified and removed according to the abnormal data elimination logic, and the data completion and standardization processing are carried out, so as to comprehensively optimize the original data.

[0093] Through a series of data processing operations, the integrity, accuracy and time sequence continuity of the health monitoring data are effectively ensured. The application of abnormal data elimination logic avoids the influence of abnormal values on the overall data. The data completion and standardization processing perfect the data content and unify the format. These processing eliminates the interference of environmental noise, load fluctuation and other factors on subsequent data analysis, and provides a reliable data basis for mechanical and electrical system state evaluation, fault diagnosis and the like based on data;

[0094] The mechanical and electrical system operation state feature extraction and multi-scale trend analysis in S4 refers to the use of a three-layer convolutional neural network with a convolution kernel of 3*3, a step of 1 and output channels of 32, 64 and 128 in sequence, to perform deep feature mining on four types of parameters of vibration, heat, electricity and fluid. For time domain vibration signals, peak value, root mean square and other features are extracted according to a specific calculation logic. The electric parameter focuses on the total harmonic distortion rate and current impact. The thermal imaging focuses on the maximum temperature difference in the high temperature area. The fluid parameter extracts the flow fluctuation rate, and realizes the integration of multi-source data features.

[0095] The extracted multi-parameter features are converted into time sequence data stream, and the operation trend of the data stream analysis equipment is analyzed. Through continuous monitoring and analysis of various parameter features, combined with the change of the key index of the root mean square of the vibration signal, the potential fault mode of the equipment is accurately identified, the operation state of the equipment is predicted in advance, and strong support is provided for the maintenance and fault warning of the mechanical and electrical system;

[0096] The historical state modeling and predictive maintenance knowledge base construction in S5 refers to the construction of a mechanical and electrical equipment state monitoring knowledge base based on the collected historical equipment state data. The support vector machine and long short-term memory network (LSTM) are used to construct a data-driven prediction model with the features of vibration root mean square and harmonic distortion rate as input. The model is deeply trained by not less than 100,000 training set data, and the prediction error is strictly controlled within 5% by using the state prediction logic optimization model of LSTM. The remaining service life and health index of the equipment are accurately output.

[0097] While completing the construction of the prediction model, a device fault case knowledge base and an operation and maintenance response strategy base are simultaneously built. The fault case knowledge base collects historical fault information to provide a reference for problem diagnosis. The operation and maintenance response strategy base formulates a targeted treatment scheme to realize the whole-process knowledge cooperation from fault prediction, diagnosis to disposal and perfect the intelligent operation and maintenance system of mechanical and electrical equipment.

[0098] In the S6, the real-time state prediction and hierarchical dynamic early warning mechanism refers to using an LSTM model to predict the trend of the future 24-hour operation state of the equipment, focusing on capturing typical fault signs such as sudden increase in vibration, distortion of current, and abnormal temperature rise in thermal imaging. Through the state prediction logic processing of the LSTM, the trend of the equipment operation is accurately deduced to provide a reliable prediction basis for fault early warning.

[0099] A three-level dynamic early warning threshold system is set: mild abnormality (prediction deviation 5%-10%), moderate abnormality (10%-15%), and serious abnormality (≥15%). The early warning generation lag time is controlled within 1 second. This mechanism combines dynamic threshold logic to automatically trigger the corresponding early warning level according to the prediction deviation. At the same time, the system supports alarm linkage and on-site inspection dispatching functions to form a closed-loop management from prediction to disposal.

[0100] In the S7, the digital twin virtual-real synchronization and whole-life cycle carbon emission mapping refers to building a digital twin platform of mechanical and electrical systems through the bidirectional fusion of finite element simulation models and real-time monitoring data. Relying on the virtual-real residual error control logic, the data synchronization quality is strictly controlled to ensure that the virtual-real data synchronization delay is ≤2 seconds and the virtual-real residual error is ≤5%, realizing the accurate mapping and dynamic linkage of the physical system and the virtual model.

[0101] Based on the real-time load, energy efficiency coefficient, and operation time of the equipment, the carbon emission of a single device is dynamically calculated, and the whole-life cycle carbon emission curve of the mechanical and electrical system is accumulated. At the same time, the platform updates the equipment health index and carbon emission prediction results in real time to realize the collaborative monitoring and management of the equipment operation state and environmental protection indicators.

[0102] In the S8, the intelligent optimization scheduling and priority operation strategy implementation refers to formulating a dynamic scheduling scheme based on the equipment health index, energy efficiency parameter, and carbon emission indicator of the digital twin platform. According to the health priority scheduling priority logic, the device start-stop sequence, load distribution, and operation priority are determined to preferentially schedule devices with good health status and high energy efficiency, laying a foundation for efficient operation of the system.

[0103] Through the above dynamic scheduling, the overall operation efficiency of the system is optimized and the carbon emission is minimized. Reasonable load distribution avoids continuous fatigue of high-load devices, prolongs the service life of the equipment, and balances efficiency and environmental protection while ensuring stable operation of the system, achieving comprehensive benefit improvement.

[0104] Among them, the cloud remote operation and maintenance monitoring and closed-loop feedback optimization in S9 refers to presenting the equipment health status, early warning information, carbon emission data and scheduling optimization results in real time through the Web terminal, APP terminal and local control terminal, synchronizing the cloud operation and maintenance decision and the field control system, following the time control logic of cloud alarm response, forming an intelligent operation and maintenance closed-loop management, and ensuring efficient connection of operation and maintenance;

[0105] Combined with new equipment operation data, the prediction model and scheduling strategy are continuously iteratively optimized, the model accuracy and strategy adaptability are continuously improved, the health management capability of the electromechanical system throughout the life cycle is gradually improved, and the whole-process efficient management from real-time monitoring, intelligent decision-making to continuous optimization is realized;

[0106] It should be noted that in this paper, relationship terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between the entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that the process, method, article or device including a series of elements not only includes those elements, but also includes other elements not explicitly listed or inherent to such process, method, article or device.

[0107] Although the embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. An artificial intelligence based electromechanical system health monitoring and optimization method, characterized in that: Comprise: S1: Based on finite element simulation model, transient simulation model and carbon emission calculation model, the digital twin architecture of electromechanical system is built, and multi-source sensor network is deployed to comprehensively cover key equipment and environmental parameters; S2: Real-time high-frequency acquisition of multi-dimensional data of electromechanical system operation, and through the multi-channel hybrid transmission architecture composed of NB-IoT, 5G wireless network and RS485 bus, the data is synchronously transmitted to the cloud or edge computing platform, ensuring the continuity, stability and low delay of data transmission; S3: The raw data collected in real time is preprocessed, and various data processing techniques such as sliding window analysis, time domain averaging and wavelet denoising are used to remove outliers and complete data completion and standardization, ensuring the completeness, accuracy and time sequence continuity of the monitoring data; S4: The cleaned multi-source data is input into a deep learning model to extract the operating state characteristics of the electromechanical system and perform multi-scale trend analysis. The deep learning model includes a three-layer convolutional neural network for mining vibration, temperature, electrical and fluid parameters, and converting the extracted multi-parameter features into time series to identify potential equipment failure modes; S5: Based on the historical data collected, a prediction model and an equipment operation and maintenance knowledge base are established. The prediction model uses a data-driven approach combining support vector machines and LSTM models, trained with at least 100,000 historical data to control the prediction error within 5%, and outputs the remaining life and health index of the equipment, while building a fault case library and an operation and maintenance strategy library; S6: Real-time prediction of future operating state of the equipment using the LSTM model, accurate deduction of the operating trend based on the fault signs of sudden vibration increase, current distortion and abnormal temperature rise in thermal imaging, setting up a three-level dynamic warning threshold system for health state monitoring, where mild abnormality corresponds to a prediction deviation of 5%~10%, moderate abnormality of 10%~15%, and severe abnormality of ≥15%, and the delay of the warning generated is controlled within 1 second to automatically trigger the corresponding level of alarm and support alarm linkage and on-site repair order; S7: Carry out digital twin virtual-real synchronization and full life cycle carbon emission mapping, build a digital twin platform through real-time monitoring data and simulation model bidirectional fusion, ensure the virtual model and physical equipment state synchronization delay ≤2 seconds, residual error ≤5% by using virtual-real residual control, and dynamically calculate the carbon emissions of a single device based on real-time load, energy efficiency coefficient and operating time, accumulate to generate the full life cycle carbon emission curve of the electromechanical system, and real-time update the equipment health index and carbon emission prediction results; S8: Based on the equipment health index, energy efficiency parameters and carbon emission indicators output by the digital twin platform, dynamically optimize the equipment scheduling scheme, adjust the equipment start-stop sequence, load distribution and operation priority, and preferentially schedule equipment with good health status and higher energy efficiency; S9: Through cloud remote operation and maintenance monitoring and closed-loop feedback optimization mechanism, the device health status, warning information, carbon emission data and scheduling optimization results are presented in real time on the Web, mobile App or local control terminal, and the cloud decision and on-site control system are synchronized to form an intelligent operation and maintenance closed loop.

2. The method of claim 1, wherein: The specific steps of the S1 multi-source sensor network construction and equipment monitoring object determination are as follows: Digital twin architecture construction and monitoring planning: relying on the finite element analysis, transient simulation and carbon emission calculation model of the building and mechanical and electrical system, the digital twin basic architecture of the mechanical and electrical system is constructed, the key monitoring objects and vulnerable parts are accurately selected through deep simulation and working condition analysis of the model, and the monitoring points are scientifically determined; Multi-source sensor deployment and measurement point optimization: According to the device characteristics and environmental parameter differences, reasonably deploy NB-IoT wireless vibration, thermal imaging, current and voltage multi-type sensors, and clearly define the performance parameter requirements of each sensor. At the same time, follow the principle of not less than 3 groups of sensing units per 100m 2 Building area configuration not less than 3 groups of sensing units, combined with multi-source sensor measurement point optimization logic, and considering the factors of building area, dynamically plan the number of measurement points; The multi-source sensor measurement point optimization expression formula is: ； In the formula, Minimum number of sensors to be deployed, Area of the monitoring zone, Ceiling function.

3. The method of claim 2, wherein: The multi-channel hybrid transmission architecture adopts an asynchronous multi-channel transmission mode combining NB-IoT, 5G wireless network and RS485 bus, and builds a wireless and wired complementary data transmission architecture, wherein the transmission delay of the NB-IoT network is not more than 2 seconds, the transmission delay of the 5G network is ≤0.5 seconds, and the transmission rate of the RS485 bus is ≥115.2 kbps. The cooperative action of each transmission channel ensures efficient and stable data upload; the data collected by different channels are synchronized after delay compensation and unified timestamp, and the multi-channel data synchronization logic satisfies the following relationship: ； In the formula, Time Synchronization data set of time, The Channel sensor delay synchronized data, The Channel transmission delay, Total number of sensor channels, multi-channel synchronization data needs to be time stamped after delay compensation.

4. The method of claim 3, wherein: In the data cleaning preprocessing process of the S3, the multi-dimensional data collected in real time are processed by combining multiple algorithms such as sliding window, time domain average and wavelet denoising, the data completion and standardization are completed while the abnormal values are eliminated, so as to improve the quality of the original monitoring data; wherein the discrimination and elimination of abnormal data are based on the following criteria: the expression formula of abnormal data elimination is: ； In the formula, Standard deviation score (Z-score), Collect data, Mean within the sliding window, Standard deviation within the sliding window, Abnormal data determination: when |Z|>3, the data is determined to be abnormal and needs to be excluded.

5. The method of claim 4, wherein: The deep learning model in the S4 adopts a three-layer convolutional neural network structure, the convolution kernel size is 3×3, the step is 1, and the output channel number is 32, 64 and 128 in sequence, so as to extract deep features of vibration, thermal imaging, electrical and fluid parameters; the time domain features of peak value and root mean square of the vibration signal are further extracted, the harmonic total distortion rate and current impact characteristics of the electrical parameters are extracted, the maximum temperature difference of the high temperature area of the thermal imaging parameters is extracted, and the flow fluctuation rate of the fluid parameters is extracted, which are used to construct a multi-parameter feature set and convert it into a time series signal to analyze the equipment operation trend; wherein the root mean square of the vibration signal is one of the key indicators of the equipment health status, and the expression formula of the root mean square of the vibration signal is: ； In the formula, RMS: root mean square value of the vibration signal, The first Vibration acceleration of the sampling point, Total number of sampling points.

6. The method of claim 5, wherein: The prediction model constructed in the step S5 is trained by combining support vector machine and LSTM model on the historical state data of the equipment, taking the features of the root mean square value of the vibration signal, the harmonic distortion rate, the high temperature point temperature of the thermal imaging and the flow fluctuation rate as input, and using not less than one hundred thousand groups of historical data samples to train the model and optimize the state prediction parameters of the LSTM model, so that the error of the predicted output of the equipment remaining service life and health index is controlled within 5%; at the same time, the equipment fault case knowledge base and operation and maintenance response strategy base are established synchronously in the process of training the prediction model, which are used to store historical fault information and corresponding disposal strategies, so as to realize the knowledge cooperation from prediction, diagnosis to maintenance decision.

7. The method of claim 6, wherein: The S6 sets the three-level dynamic early warning threshold system to realize the hierarchical alarm of the predicted state, wherein the mild abnormality corresponds to a deviation of 5-10% between the predicted value and the current actual value, the moderate abnormality corresponds to a deviation of 10-15%, and the serious abnormality corresponds to a deviation of ≥15%; the corresponding level of alarm signal is automatically triggered according to the prediction deviation through the dynamic threshold logic, and the lag time of the early warning is limited within 1 second; The expression formula of the three-level dynamic early warning threshold is: ； In the formula, a prediction error percentage, a predicted state value, a current actual monitored state value; Warning level: mild abnormality ; moderate abnormality ; severe abnormality .

8. The method of claim 7, wherein: In the S7, in order to ensure the synchronization accuracy of the virtual model and the physical device, the virtual-real residual error control strategy is adopted to strictly control the data synchronization quality of the digital twin platform, to ensure that the virtual-real data synchronization delay is not more than 2 seconds and the virtual-real data residual error is not more than 5%, to realize the accurate mapping and dynamic linkage of the physical system state on the digital twin platform; based on the real-time load, energy efficiency coefficient and running time data of the device obtained by the digital twin platform, the carbon emission of a single device is dynamically calculated, and the full life cycle carbon emission curve of the electromechanical system is accumulated and generated; The expression formula of the digital twin virtual-real residual error control is: ； In the formula, virtual-real residual percentage, digital twin virtual state data, real-time actual monitoring data.

9. The method of claim 8, wherein: In the S8, the health priority scheduling strategy is adopted, the operation priority score of each device is calculated according to the parameters of the device health index, the current energy efficiency coefficient and the unit time carbon emission, and the device with good health status, high energy efficiency and low carbon emission is preferentially started to undertake the operation load; by dynamically adjusting the start-stop sequence and load distribution of each device, the optimal overall operation efficiency and minimization of carbon emission of the electromechanical system are realized, and the long-term overloading of a single device is avoided to accelerate the aging, and the safety and economy of the system operation are improved; The expression formula of the health priority scheduling priority is: ； In the formula, Device operational priority score, Device health index (0-1), Device current energy efficiency coefficient (0-1), Device current carbon emission per unit time.

10. The method of claim 9, wherein: In the S9, the time control logic of cloud alarm response is followed, the delay management of the collaborative process of device remote monitoring and on-site control is carried out, the total response time from data uploading, cloud processing to client pushing of the alarm signal is controlled within 3 seconds; the total delay of the alarm response satisfies the following relationship; The expression formula of the cloud alarm response delay is: ； In the formula, Alert response total delay, Data upload transmission delay, Cloud processing calculation delay, Cloud to client alarm push delay, ensure Ttotal≤3 seconds, meet the real-time alarm response requirements of the cloud.

Images