Amorphous alloy motor operation efficiency monitoring method and system based on artificial intelligence

Abstract

The application discloses an amorphous alloy motor operation efficiency monitoring method and system based on artificial intelligence, relates to the technical field of electrical engineering, and comprises a plurality of function modules, including: a multi-dimensional data acquisition module, which collects motor operation multi-field coupling parameters and environmental working condition data; the motor operation multi-field coupling parameters include electrical coupling parameters, thermal coupling parameters and mechanical coupling parameters; an edge computing module, which pre-processes the motor operation multi-field coupling parameters and environmental working condition data, introduces multi-field coupling characteristics, extracts multi-dimensional characteristics and multi-sensor time sequence characteristics; and a cloud artificial intelligence analysis module, which, based on historical motor operation data, adopts an improved LSTM and Attention mechanism, fuses a Transformer architecture, constructs a dynamic mapping model, inputs multi-dimensional characteristics into the dynamic mapping model, outputs an energy efficiency grade, and fuses multi-sensor time sequence characteristics based on the energy efficiency grade by using a graph neural network.

Description

Technical Field

[0001] This invention relates to the field of electrical engineering technology, specifically to a method and system for monitoring the operating performance of amorphous alloy motors based on artificial intelligence. Background Technology

[0002] Amorphous alloy motors are a new type of high-efficiency motor that uses amorphous alloys, also known as "metallic glass," as a substitute for traditional silicon steel sheets as the core material. Their most significant characteristic is extremely low core loss, resulting in ultra-high efficiency and significant energy savings. They are made by cooling molten metal at ultra-high speeds of millions of degrees per second. Because the atoms do not have time to arrange themselves into an ordered crystal structure, they ultimately form an amorphous structure similar to glass, characterized by "long-range disorder and short-range order." Applications include high-efficiency industrial motors such as pumps, fans, and compressors in continuously operating equipment. Although amorphous alloy motors possess advantages such as high efficiency and low temperature rise, their actual operating status is still affected by multiple factors, including load fluctuations, environmental conditions, material aging, and assembly stress. To fully unleash their performance potential, refined perception and dynamic evaluation of operating efficiency are necessary. Data-driven intelligent monitoring methods can extract key features from massive amounts of operating data, enabling accurate judgment of motor energy efficiency, health status, and lifespan trends. This provides a key technological path for predictive maintenance, operational optimization, and continuous energy efficiency improvement.

[0003] Existing motor monitoring systems have significant shortcomings in adaptability, real-time performance, early warning capabilities, and optimization capabilities. They have not developed dedicated evaluation models for the unique material characteristics of amorphous alloy motors, such as "low loss, high-frequency response, and stress sensitivity," resulting in insufficient adaptation between monitoring and actual operating conditions. Performance evaluation relies heavily on theoretical calculations or offline test data, making it difficult to dynamically estimate motor operating efficiency in real time and failing to reflect the true performance status. Fault diagnosis is mostly reactive, lacking early warning capabilities for minor performance degradation of motors and easily missing the fault intervention window. Furthermore, the lack of adaptive optimization mechanisms prevents dynamic adjustment of operating strategies based on load fluctuations and environmental changes, making it difficult to continuously maintain the optimal energy efficiency level of the motor. Summary of the Invention

[0004] (a) Technical problems to be solved

[0005] To address the shortcomings of existing technologies, this invention provides an artificial intelligence-based method and system for monitoring the operating efficiency of amorphous alloy motors. This solves the problems in existing technologies, such as the lack of a dedicated evaluation model for the characteristics of amorphous alloys, the lag in real-time dynamic efficiency estimation and fault diagnosis, the difficulty in timely early warning, and the inability to dynamically adjust operating strategies according to load and environment to maintain optimal energy efficiency.

[0006] (II) Technical Solution

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] An AI-based amorphous alloy motor performance monitoring system includes:

[0009] The multi-dimensional data acquisition module collects multi-field coupling parameters of motor operation and environmental operating condition data, including electrical coupling parameters, thermal coupling parameters and mechanical coupling parameters.

[0010] The edge computing module preprocesses the multi-field coupling parameters of the motor operation and environmental condition data, and introduces multi-field coupling features to extract multi-dimensional features and multi-sensor time-series features.

[0011] The cloud-based artificial intelligence analysis module, based on historical motor operation data, adopts an improved LSTM and Attention mechanism and integrates the Transformer architecture to build a dynamic mapping model. Multi-dimensional features are input into the dynamic mapping model, and the energy efficiency level is output. Based on the energy efficiency level, graph neural networks are used to fuse time-series features from multiple sensors to output the motor health index and motor remaining life prediction.

[0012] The human-computer interaction decision module visualizes the motor health index and the prediction of the motor's remaining lifespan using a digital twin, and outputs a 3D dynamic coupled display of the motor.

[0013] Furthermore, the acquisition of multi-field coupling parameters of the motor operation and environmental condition data includes:

[0014] Electrical coupling parameters include three-phase voltage, current, motor power, power factor, and rotational frequency;

[0015] Thermal coupling parameters include stator winding temperature, bearing temperature, and infrared thermal image of the housing;

[0016] Mechanical coupling parameters include triaxial vibration acceleration and acoustic noise;

[0017] Environmental operating data includes ambient temperature and humidity, cooling fan speed, load torque, number of start-stop cycles, and operating duration.

[0018] Furthermore, the specific process of introducing multi-field coupling features is as follows:

[0019] Preprocessing steps include cleaning, synchronizing, and standardizing the multi-field coupling parameters of motor operation and environmental condition data;

[0020] By acquiring preprocessed multi-field coupling parameters of motor operation and environmental condition data, an electrical-thermal coupling factor and a mechanical-electrical coupling factor are constructed. The electrical-thermal coupling factor is used to determine whether the electrical or thermal system is abnormal during motor operation; the mechanical-electrical coupling factor is used to determine the degree of coupling between mechanical vibration and electrical harmonics during motor operation, and whether there are potential faults in its mechanical or electrical components.

[0021] Furthermore, the multidimensional feature extraction process is as follows:

[0022] After extracting multi-field coupling features, multi-dimensional features are extracted, including electrical dimension features, thermal dimension features, heat flux density distribution features, mechanical dimension features, and acoustic noise features.

[0023] The multi-sensor temporal feature extraction process is as follows:

[0024] Based on the extraction of multidimensional features, the variation law of multidimensional features over time is analyzed, and multi-sensor time-series features are extracted, including time-series statistical features, time-series trend features, and time-series abrupt change features.

[0025] Furthermore, the improved LSTM and Attention mechanism obtains historical motor operation data from an existing database and preprocesses it. The preprocessed historical motor operation data is then input into the LSTM network in a time series manner. Long-term information is selectively stored through memory units to capture the long-term dependencies of motor operation data in the time dimension. The Attention mechanism is applied to the output of the LSTM, using the LSTM output as the input of the query, key, and value. Attention weights are obtained by calculating the similarity between the query and the key.

[0026] Furthermore, the fused Transformer architecture includes encoder construction and positional encoding;

[0027] The encoder is constructed using the Transformer encoder structure, which includes a multi-head self-attention mechanism and a feedforward neural network. The attention weights obtained by LSTM and the Attention mechanism are input into the Transformer encoder. The multi-head self-attention mechanism maps the input attention weights to different subspaces, generates multiple Query, Key, and Value matrices to capture global dependencies, and concatenates the results of each head for linear transformation. The feedforward neural network enhances the model's learning ability.

[0028] Positional encoding, typically created using sine and cosine functions, is added to the input attention weights to assign unique information about the position of each attention weight.

[0029] Furthermore, the process of setting the output energy efficiency level is as follows:

[0030] The feature sequence is first modeled by an improved LSTM to model long-term temporal dependencies, preserving dynamic features of energy efficiency assessment; then it is weighted by an Attention mechanism to highlight key variables; subsequently, it is fed into a Transformer encoder to capture global interactions of multiple variables and enhance feature discriminativeness; through a feature fusion layer, the temporal evolution trend of the LSTM and the global association of the Transformer are deeply integrated to form a comprehensive representation of the input classification layer, which outputs the probability distribution of IE4 and IE5 energy efficiency levels, and the level with the highest probability is taken as the judgment result.

[0031] Furthermore, the process of predicting the output motor health index and remaining motor lifespan is as follows:

[0032] We acquire energy efficiency levels and multi-sensor time-series features to form a basic dataset; we construct a graph structure data with multi-sensor nodes and physical field coupling relationships as edges; we globally pool the node features of the last layer of the GNN and input them into a fully connected layer; we comprehensively consider the energy efficiency level decay trend, the anomaly degree of multi-sensor and the stability of coupling relationships, and obtain a health index between 0 and 1 by weighted summation.

[0033] Using the time series of health index as input, combined with the historical change curve of energy efficiency level, a long short-term memory network is used to fit the health state decay model, learn the evolution law to establish the mapping between the running time, and predict the remaining lifespan by combining the current energy efficiency level decay rate.

[0034] Furthermore, the specific process of the 3D dynamic coupling display of the output motor is as follows:

[0035] The system integrates motor health index, motor remaining life prediction, and multi-field coupling parameters and environmental conditions of motor operation into a digital twin system. It also identifies the corresponding components of the motor's 3D model with differentiated visualizations and dynamically presents countdown information based on the motor's remaining life prediction. The system integrates simulation algorithms to map the dynamic changes of multi-field coupling parameters and environmental conditions of motor operation onto the model. Finally, it uses graphics rendering and animation technology to fuse information and form a 3D interactive interface.

[0036] The method for monitoring the operating performance of amorphous alloy motors based on artificial intelligence includes the following steps:

[0037] Step 1: Collect multi-field coupling parameters of motor operation and environmental condition data. The multi-field coupling parameters of motor operation include electrical coupling parameters, thermal coupling parameters and mechanical coupling parameters;

[0038] Step 2: Preprocess the multi-field coupling parameters of the motor operation and the environmental condition data, and introduce multi-field coupling features to extract multi-dimensional features and multi-sensor time-series features;

[0039] Step 3: Based on historical motor operation data, an improved LSTM and Attention mechanism is adopted and integrated with the Transformer architecture to build a dynamic mapping model. Multi-dimensional features are input into the dynamic mapping model to output the energy efficiency level. Based on the energy efficiency level, graph neural network is used to fuse time-series features from multiple sensors to output the motor health index and motor remaining life prediction.

[0040] Step 4: Visualize the motor health index and the motor remaining life prediction using digital twins, and output a 3D dynamic coupling display of the motor.

[0041] (III) Beneficial Effects

[0042] This invention provides a method and system for monitoring the operating performance of amorphous alloy motors based on artificial intelligence, which has the following beneficial effects:

[0043] (1) In this solution, monitoring the operating efficiency of amorphous alloy motors using artificial intelligence is a key technical path to achieve predictive maintenance, optimize operating efficiency, extend equipment life and reduce energy consumption. Although amorphous alloy motors are highly efficient and have low temperature rise, their operating status is still affected by various factors such as load, environment, aging, and assembly stress. AI can extract patterns from massive amounts of data to achieve accurate monitoring.

[0044] (2) This solution takes advantage of the unique material characteristics of amorphous alloy motors, such as low loss, high frequency response, and stress sensitivity, and constructs multi-field coupling features and incorporates a dynamic mapping model to overcome the limitations of the general model of traditional monitoring systems and the insufficient adaptability of amorphous alloy motors. Compared with the method of relying on theoretical values ​​or offline testing, it can more accurately capture the correlation between material characteristics and operating status, significantly reduce the performance evaluation error, and provide reliable support for the special state analysis of amorphous alloy motors.

[0045] (3) This solution uses graph neural networks to fuse multi-sensor time-series features and energy efficiency level data to construct a health index assessment system, which can keenly capture the small performance degradation signals of motors. Compared with the traditional lagging fault diagnosis mode, it can advance the fault warning node to the early stage of performance degradation. Combined with the remaining life prediction, it enables operation and maintenance to carry out predictive maintenance instead of post-maintenance maintenance, effectively reducing unplanned downtime losses. Attached Figure Description

[0046] Figure 1 This is a schematic diagram of the system flow of the present invention;

[0047] Figure 2 This is a schematic diagram of the overall method of the present invention. Detailed Implementation

[0048] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0049] Example 1:

[0050] Please see Figure 1 This embodiment provides an artificial intelligence-based amorphous alloy motor operating performance monitoring system, which includes:

[0051] Amorphous alloy motors are a new type of high-efficiency motor that uses amorphous alloys instead of traditional silicon steel sheets as the core material. They are primarily used in the stator, with some designs also applying to the rotor. This material is produced by rapidly cooling molten iron-based alloys at a rate on the order of millions of degrees per second, preventing atoms from forming an ordered lattice and resulting in an amorphous structure characterized by "long-range disorder and short-range order." This unique structure significantly reduces core losses compared to traditional materials, resulting in a substantial increase in motor efficiency and outstanding energy savings.

[0052] Despite the inherent advantages of amorphous alloy motors, such as high efficiency and low temperature rise, their actual operating conditions are still affected by multiple factors, including load fluctuations, environmental conditions, material aging, and assembly stress. To fully unleash their performance potential, refined perception and dynamic evaluation of operational efficiency are necessary. Data-driven intelligent monitoring methods can extract key features from massive amounts of operational data, enabling accurate assessment of motor energy efficiency, health status, and lifespan trends. This is a crucial technological path supporting predictive maintenance, operational optimization, and continuous energy efficiency improvement.

[0053] The multi-dimensional data acquisition module collects multi-field coupling parameters of motor operation and environmental operating condition data, including electrical coupling parameters, thermal coupling parameters and mechanical coupling parameters.

[0054] The multi-field coupling parameters of the electrical system are the core dynamic data for motor operation. The electrical coupling parameters can reflect the complex interaction between the motor and the electrical system and other physical fields, such as thermal and mechanical fields.

[0055] Electrical coupling parameters, such as three-phase voltage and current, including harmonic components, motor power, power factor, rotational frequency, voltage and current phase difference, voltage modulation depth, etc., reflect the interaction between the motor electrical system and other physical fields. For example, changes in voltage and current will affect the power output of the motor, and thus become related to the thermal field and mechanical field.

[0056] Motor power includes active power and apparent power.

[0057] For example, in amorphous alloy motors, due to the special magnetic properties of the material, characteristics such as hysteresis and eddy currents can cause changes in the phase relationship between voltage and current. By acquiring the phase difference in real time, changes in electrical characteristics caused by variations in magnetic properties can be accurately detected. For instance, when an abnormality occurs in the motor's magnetic circuit, the phase difference will deviate from the normal range. Voltage modulation depth The modulation depth is a parameter related to the inverter's output voltage, defined as the ratio of the modulation voltage to the carrier voltage. The modulation depth affects the quality of the motor's input power. An inappropriate modulation depth can lead to waveform distortion of the motor's input voltage, increase harmonic content, and thus affect the motor's operating efficiency. It can also cause the motor to generate more losses, which is reflected in the thermal parameters as abnormal temperature changes.

[0058] Thermal coupling parameters include stator winding temperature, multiple monitoring points, bearing temperature, infrared thermography of the casing, and newly added heat flux density. Electrical and mechanical losses during motor operation are converted into heat. These thermal parameters reflect the distribution and transfer of heat in various parts of the motor, and changes in heat also affect the performance of the electrical and mechanical components.

[0059] Mechanical coupling parameters include triaxial vibration acceleration, acoustic noise, and mechanical strain rate. The mechanical operating state of the motor, such as vibration and strain, is closely related to electrical drive and thermal changes. For example, electrical faults may cause abnormal vibrations, while excessively high temperatures may cause deformation of mechanical components.

[0060] Environmental condition data collection:

[0061] In addition to the multi-field coupling parameters of motor operation, the multi-dimensional data acquisition module also collects environmental operating condition data. By reflecting the external environment and working conditions of the motor during operation, it combines the multi-field coupling parameters of the motor, including ambient temperature and humidity, cooling wind speed, load torque, number of start-stop cycles and running time.

[0062] Ambient temperature and humidity: Ambient temperature and humidity affect the heat dissipation and insulation performance of the motor. Excessively high ambient temperature reduces the motor's heat dissipation efficiency, leading to increased motor temperature and consequently affecting electrical and mechanical performance. Excessively high ambient humidity may cause moisture to accumulate on the motor's insulating components, reducing insulation resistance and increasing the risk of electrical failures.

[0063] Cooling airflow: Cooling air is used to remove the heat generated by the motor during operation, and the cooling airflow directly affects the motor's heat dissipation effect. Insufficient cooling airflow will cause heat to accumulate in the motor, causing the temperature to rise, thus affecting the motor's operating efficiency and lifespan; excessive cooling airflow may increase the fan's energy consumption and may also have a certain impact on the motor's mechanical structure.

[0064] Load torque: Load torque reflects the motor's workload. Changes in load torque will cause changes in the motor's electrical parameters such as current and power, and will also affect the motor's mechanical vibration and heat generation. Equivalent power is an equivalent expression of load torque and also reflects the motor's workload.

[0065] Number of start-stop cycles and running time: Too many start-stop cycles will cause frequent impacts on the electrical and mechanical components of the motor, affecting its service life; running time reflects the cumulative working time of the motor. Long-term operation will cause the heat generated by the motor to accumulate, the temperature to rise, and the wear of mechanical components will also be aggravated with the increase of running time.

[0066] The edge computing module preprocesses the multi-field coupling parameters of the motor operation and environmental condition data, and introduces multi-field coupling features to extract multi-dimensional features and multi-sensor time-series features.

[0067] The multi-field coupling parameters of the motor operation and environmental condition data are preprocessed. The preprocessing operations include data cleaning to remove outliers and noise, and to remove jump data caused by sensor failures and stray data caused by electromagnetic interference. Data synchronization is also performed, as different sensors may have different sampling frequencies. The data from different sensors are aligned in the time dimension to ensure the temporal consistency of subsequent analysis. Data standardization is also performed to convert data of different dimensions and ranges into a unified range.

[0068] Introducing multi-field coupling features:

[0069] During motor operation, electrical, thermal, and mechanical physical fields interact and couple with each other. For example, changes in the power of the electrical system affect the heat generated by the motor, thereby altering the temperature distribution of the thermal field; mechanical vibrations, in turn, are associated with harmonics in the electrical system, affecting the motor's operating state. Analyzing the parameters of a single electrical, thermal, or mechanical physical field in isolation is insufficient to comprehensively and accurately reflect the overall operating state of the motor.

[0070] By constructing an electrical-thermal coupling factor, the efficiency of the motor in converting electrical energy into heat energy can be reflected, linking parameters such as power and heat flux density. The coupling factor indicates whether the conversion of electrical energy to heat energy is normal; if the conversion efficiency is abnormal, there may be a problem with the electrical or thermal system.

[0071] The mechanical-electrical coupling factor reflects the degree of coupling between mechanical vibration and electrical harmonics, linking parameters such as vibration acceleration and current harmonic distortion rate. It determines whether the interaction between mechanical vibration and electrical harmonics is normal; if the coupling is abnormal, there may be potential faults in the mechanical or electrical components of the motor.

[0072] By constructing multi-field coupling features, the interaction between different physical fields is captured, laying the foundation for subsequent analysis of the motor's operating status.

[0073] Multidimensional feature extraction:

[0074] The coupling factor after multi-field coupling features are extracted from multi-dimensional features. Key information is extracted from different dimensions such as electrical, thermal, and mechanical to form multi-dimensional features, including electrical dimension features, thermal dimension features, heat flux density distribution features, mechanical dimension features, and acoustic noise features.

[0075] Electrical dimension characteristic: Harmonic distortion rate (THD), calculated as the ratio of the effective value of voltage or current harmonic components to the effective value of the fundamental frequency, reflects the degree of distortion of the electrical waveform. The formula is:

[0076] Taking current as an example, The effective value of the current for the nth harmonic is... is the effective value of the fundamental current. Excessive harmonic distortion rate may indicate a fault in the electrical system; n is the harmonic order, n=2,3,4,...

[0077] Power factor The ratio of active power to apparent power reflects the degree to which a motor effectively utilizes electrical energy. Where P is active power and S is apparent power.

[0078] Thermal dimension characteristics: Temperature rise slope: Calculates the rate of temperature increase of the motor per unit time, reflecting the speed of heat accumulation in the motor. For example, over a period of time... Inside, the temperature from Rise to The slope of the temperature rise is An abnormal temperature rise slope may indicate a problem with the heat dissipation system or internal losses.

[0079] Heat flux density distribution characteristics: By analyzing the distribution of heat flux density in different parts of the motor, the maximum, minimum and average values ​​are extracted to understand the unevenness of heat transfer.

[0080] Mechanical Dimension Features: Vibration Spectrum Features: Fourier transform is performed on the triaxial vibration acceleration signal to obtain the vibration spectrum, and features such as peak frequency and spectral energy distribution are extracted. Different mechanical faults, such as bearing wear and rotor eccentricity, will produce peaks at specific frequencies, and these features can be used to identify the fault type.

[0081] Acoustic noise characteristics: By analyzing acoustic noise signals, features such as frequency, amplitude, and spectral entropy of the noise can be extracted. Changes in noise can reflect the operating status of the mechanical components of the motor.

[0082] Multi-sensor temporal feature extraction:

[0083] Multi-sensor temporal feature extraction, based on the extraction of multi-dimensional features, analyzes the changing patterns of features over time to extract multi-sensor temporal features, including temporal statistical features, temporal trend features, and temporal abrupt change features.

[0084] Time series statistical features obtain the mean and variance, calculate the mean and variance of a feature within a sliding time window, and reflect the central tendency and dispersion of the feature.

[0085] For example, using a 10-second sliding window, the mean and variance of vibration acceleration within the window are calculated. Changes in the mean may indicate the trend of vibration intensity, while changes in the variance reflect the stability of the vibration. Maximum and minimum values: Extracting the maximum and minimum values ​​of the features within the sliding window helps identify extreme cases. Extreme values ​​may correspond to abnormal events in the motor, such as a sudden increase in current due to a short-term overload.

[0086] The time-series trend feature is analyzed by linear regression to determine the linear change trend of the feature over time. The formula is y=a+bt, where y is the feature value, t is time, a is the intercept, and b is the slope. A positive slope indicates that the feature is in an upward trend, and a negative slope indicates that it is in a downward trend.

[0087] For example, a positive linear trend with a large slope in stator winding temperature may indicate a problem of continuous overheating in the motor. Periodic features can be extracted using Fourier transform or autocorrelation analysis. Motor operation may exhibit periodic vibrations or current fluctuations related to speed; identifying periodic features can determine whether the motor is operating according to a normal pattern.

[0088] Temporal abrupt change features are detected by employing methods such as the Cumulative Sum CUSUM algorithm or Bayesian change point detection to identify abrupt changes in features over time. For example, when a motor malfunctions, such as a sudden bearing jamming, features like vibration acceleration or current will exhibit abrupt changes; abrupt change point detection can promptly identify these anomalies.

[0089] The cloud-based artificial intelligence analysis module, based on historical motor operation data, adopts an improved LSTM and Attention mechanism and integrates the Transformer architecture to build a dynamic mapping model. Multi-dimensional features are input into the dynamic mapping model, and the energy efficiency level is output. Based on the energy efficiency level, graph neural networks are used to fuse time-series features from multiple sensors to output the motor health index and motor remaining life prediction.

[0090] Improved LSTM and Attention Mechanism:

[0091] The LSTM architecture consists of input gates, forget gates, output gates, and memory units. It acquires and preprocesses historical motor operation data from an existing database, then inputs this preprocessed data into the LSTM network in a time-series format. The memory units selectively store long-term information, addressing the vanishing and exploding gradient problems inherent in traditional recurrent neural networks and capturing the long-term dependencies of motor operation data over time. For example, the LSTM can retain the slow changing trends of parameters such as temperature and current during long-term motor operation by storing this information over time.

[0092] By introducing an attention mechanism, this approach is applied to the output of the LSTM. The LSTM output serves as the input for the query, key, and value. Attention weights are obtained by calculating the similarity between the query and key, typically using a dot product or other distance metrics. These weights represent the importance of the input data at different time steps when calculating the current output. For example, under certain operating conditions, the motor's current parameters have a more critical impact on performance. The attention mechanism would assign higher weights to current-related data, highlighting the temporal characteristics of parameters that significantly affect performance, allowing the model to focus more intently on key information.

[0093] Integrating Transformer architecture:

[0094] The Transformer architecture includes encoder construction and positional encoding.

[0095] The encoder is constructed using the Transformer encoder architecture, with its core components being a multi-head attention mechanism and a feedforward neural network. Data processed by LSTM and the attention mechanism is input into the Transformer encoder. In the multi-head attention mechanism, the input data is linearly projected into multiple different subspaces, generating multiple Query, Key, and Value matrices. Attention is calculated for each subspace, and the results from each head are concatenated and linearly transformed to obtain the output of the multi-head attention mechanism. The encoder captures global dependencies between operating parameters from different representation subspaces. For example, during motor operation, the correlations between electrical, thermal, and mechanical parameters are observed, regardless of their temporal distance; the Transformer can learn these relationships.

[0096] Positional encoding is used to encode the input data so that the model can perceive the temporal order of the data, since the Transformer architecture itself does not include modeling of data order. Typically, sine and cosine functions are used to generate positional encoding vectors, which are then added to the input data to assign unique positional information to each location of the input data.

[0097] The process of obtaining energy efficiency ratings:

[0098] The feature sequence is fed into the improved LSTM module. LSTM, with its gating mechanism, can effectively model the long-term temporal dependencies of parameters such as current and temperature during motor operation. The input gate filters currently valid information, the forget gate removes redundant historical states, and the output gate regulates the release of memory cells, preserving dynamic features that are discriminative for energy efficiency assessment.

[0099] The LSTM output is weighted using an attention mechanism. By calculating the correlation between features, weights are dynamically allocated to highlight variables that have a more significant impact on energy efficiency. For example, under high load conditions, current harmonic distortion rate may play a dominant role in efficiency; in this case, its weight will be automatically increased, allowing the model to focus on key criteria and improve judgment accuracy.

[0100] The weighted features are fed into the Transformer encoder. Its multi-head self-attention structure maps the features to multiple subspaces, modeling global interactions between different dimensions—capturing cross-domain couplings such as voltage fluctuations and vibration responses regardless of temporal distance. The subsequent feedforward network further enhances the nonlinear expressive power and improves feature discriminativeness.

[0101] Information is not simply transferred between modules, but is deeply integrated through feature fusion layers, such as fully connected layers. The temporal evolution trend depicted by LSTM and the multivariate global correlation captured by Transformer are co-encoded, forming a comprehensive representation that combines temporal dynamics and multiphysics coupling characteristics.

[0102] The comprehensive representation is input into the classification layer, which is usually a fully connected layer with a Softmax function. Based on the preset energy efficiency level, such as IE4, IE5, and Super IE5, the probability distribution of each level is output, and the level corresponding to the highest probability is used as the judgment result of the current motor operating energy efficiency.

[0103] Motor health index and prediction of remaining motor life:

[0104] S301: Integrating Basic Data and Features. The energy efficiency level output by the dynamic mapping model is used as a key reference indicator and integrated with multi-sensor time-series features extracted by the edge computing module, such as harmonic variation trends of electrical parameters, dynamic temperature rise curves of thermal parameters, and vibration time-series features of mechanical parameters. This integrates multi-dimensional dynamic information about motor operation, where the energy efficiency level reflects the overall performance status, and the multi-sensor time-series features reflect the real-time changes in various physical fields, together forming the basic dataset for model input.

[0105] S302: Constructing Graph-Structured Data. Multi-sensor nodes, such as voltage sensors, temperature sensors, and vibration sensors, are used as nodes in the graph. Node characteristics include the temporal features of the corresponding sensor and its correlation with energy efficiency levels, determined by calculating the correlation coefficient between sensor data and changes in energy efficiency levels. Edges between nodes are constructed based on physical field coupling relationships. For example, the edge weight between an electrical sensor and a thermal sensor is determined by the degree of influence of changes in electrical parameters on the temperature field, based on the influence coefficient statistically derived from historical data. The edge weight between a mechanical sensor and an electrical sensor is set according to the coupling strength between vibration characteristics and current harmonics. In this way, unstructured time-series data is transformed into structured graph data to capture the complex relationships between sensors.

[0106] S303: Output Health Index HI. Global pooling is performed on the node features of the last layer of the GNN to obtain the overall fused feature vector of the motor, which is then input into the fully connected layer to calculate the health index. The calculation of the health index comprehensively considers the energy efficiency level degradation trend, such as the rate of decline from above IE5 to IE5, the degree of anomalies in multiple sensors, such as the magnitude of vibration characteristics deviating from the normal range, and the stability of coupling relationships, such as the fluctuation degree of the electrical-thermal coupling factor. A weighted summation is used to obtain a value between 0 and 1, where 1 represents the best health condition and 0 represents near failure. The weight coefficients are calibrated based on historical fault data using a logistic regression model to ensure sensitivity to key fault precursors.

[0107] S304: Predicting Remaining Lifetime (RUL). Using the time series of the health index as input, combined with historical energy efficiency rating curves, a Long Short-Term Memory (LSTM) network is employed to fit a health status decay model. By learning the evolutionary pattern of "health index decline—energy efficiency rating reduction—final failure" from historical data, a mapping relationship between the health index and operating time is established. For example, if the current health index is 0.7, and the model identifies a decay trend similar to a historical failure case—which took 1200 hours to decay from 0.7 to the failure threshold of 0.1—then the remaining lifetime is predicted by combining the current energy efficiency rating decay rate. Simultaneously, confidence interval calculation is introduced, and Monte Carlo simulations are used to output RUL predictions under different probabilities, such as an RUL of 1000-1100 hours at a 90% confidence level, improving the reliability of the prediction results.

[0108] The human-computer interaction decision module visualizes the motor health index and the prediction of the motor's remaining lifespan using a digital twin, and outputs a 3D dynamic coupled display of the motor.

[0109] The motor health index, remaining life prediction results, and real-time multiphysics data, including electrical parameters such as voltage and current harmonics, thermal parameters such as temperature distribution, and mechanical parameters such as vibration amplitude, are all integrated into the digital twin system. Based on the numerical range of the health index, corresponding components in the motor's 3D model are assigned differentiated visual labels: green for good health and red for deteriorating health, enabling intuitive assessment of the health status. Simultaneously, combined with the remaining life prediction results, countdown information such as "Remaining Life: 500h" is dynamically displayed in prominent positions on the model, or a progress bar visually reflects the proportion of remaining life to the designed lifespan, helping users quickly grasp the equipment's usable time.

[0110] In terms of multiphysics coupling visualization, the system integrates professional simulation algorithms to map the dynamic changes of electrical, thermal, and mechanical parameters onto a 3D model. For example, when current harmonic distortion occurs, the winding area displays the harmonic distribution in real time through flowing colored lines or light and shadow effects; the temperature field uses color gradient mapping—the low-temperature region is blue, transitioning to yellow, orange, and red as the temperature rises, clearly showing the heat distribution and heat conduction path; when abnormal vibration occurs, relevant components, such as bearings, will simulate dynamic shaking according to the actual frequency and amplitude, intuitively reflecting the abnormal mechanical state.

[0111] Utilizing graphics rendering and animation technologies, the aforementioned health indicators, lifespan warnings, and multiphysics dynamic simulations are integrated into a unified 3D interactive interface. Users can observe the overall machine operating status from any perspective, or focus on specific components to view details. This visualization system not only achieves an integrated presentation of motor health status, remaining lifespan, and multiphysics coupling effects, but also provides intuitive and reliable technical support for operation and maintenance decisions, fault warnings, and energy efficiency optimization.

[0112] Example 2:

[0113] Please see Figure 2 Based on Example 1, this embodiment also provides an artificial intelligence-based method for monitoring the operating performance of amorphous alloy motors, including the following specific steps:

[0114] Step 1: Collect multi-field coupling parameters of motor operation and environmental condition data. The multi-field coupling parameters of motor operation include electrical coupling parameters, thermal coupling parameters and mechanical coupling parameters;

[0115] Step 2: Preprocess the multi-field coupling parameters of the motor operation and the environmental condition data, and introduce multi-field coupling features to extract multi-dimensional features and multi-sensor time-series features;

[0116] Step 3: Based on historical motor operation data, an improved LSTM and Attention mechanism is adopted and integrated with the Transformer architecture to build a dynamic mapping model. Multi-dimensional features are input into the dynamic mapping model to output the energy efficiency level. Based on the energy efficiency level, graph neural network is used to fuse time-series features from multiple sensors to output the motor health index and motor remaining life prediction.

[0117] Step 4: Visualize the motor health index and the motor remaining life prediction using digital twins, and output a 3D dynamic coupling display of the motor.

[0118] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0119] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0120] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. An artificial intelligence-based amorphous alloy motor operation performance monitoring system, characterized in that: The system includes: The multi-dimensional data acquisition module collects multi-field coupling parameters of motor operation and environmental operating condition data, including electrical coupling parameters, thermal coupling parameters and mechanical coupling parameters. The edge computing module preprocesses the multi-field coupling parameters of the motor operation and environmental condition data, and introduces multi-field coupling features to extract multi-dimensional features and multi-sensor time-series features. The cloud-based AI analysis module, based on historical motor operation data, employs an improved LSTM and Attention mechanism and integrates the Transformer architecture to construct a dynamic mapping model. Multi-dimensional features are input into the dynamic mapping model, which outputs the energy efficiency level and the probability distribution of IE4 and IE5 within the energy efficiency level. The level corresponding to the highest probability is taken as the judgment result. Based on the energy efficiency level, a graph structure data is constructed with multiple sensors as nodes and physical field coupling relationships as edges. The graph neural network is used to fuse the time-series features of multiple sensors to output the motor health index and the prediction of the motor's remaining lifespan. The human-computer interaction decision module visualizes the motor health index and the prediction of the motor's remaining lifespan using a digital twin, and outputs a 3D dynamic coupled display of the motor.

2. The artificial intelligence-based amorphous alloy motor operation performance monitoring system according to claim 1, characterized in that, The collected motor operation multi-field coupling parameters and environmental condition data include: Electrical coupling parameters include three-phase voltage, current, motor power, power factor, and rotational frequency; Thermal coupling parameters include stator winding temperature, bearing temperature, and infrared thermal image of the housing; Mechanical coupling parameters include triaxial vibration acceleration and acoustic noise; Environmental operating data includes ambient temperature and humidity, cooling fan speed, load torque, number of start-stop cycles, and operating duration.

3. The artificial intelligence-based amorphous alloy motor operation performance monitoring system according to claim 1, characterized in that, The specific process of introducing multi-field coupling features is as follows: Preprocessing steps include cleaning, synchronizing, and standardizing the multi-field coupling parameters of motor operation and environmental condition data; By acquiring preprocessed multi-field coupling parameters of motor operation and environmental condition data, an electrical-thermal coupling factor and a mechanical-electrical coupling factor are constructed. The electrical-thermal coupling factor is used to determine whether the electrical or thermal system is abnormal during motor operation. The mechanical-electrical coupling factor is used to determine the degree of coupling between mechanical vibration and electrical harmonics during motor operation, and whether there are potential faults in its mechanical or electrical components.

4. The artificial intelligence-based amorphous alloy motor operation performance monitoring system according to claim 2, characterized in that: The multidimensional feature extraction process is as follows: After extracting multi-field coupling features, multi-dimensional features are extracted, including electrical dimension features, thermal dimension features, heat flux density distribution features, mechanical dimension features, and acoustic noise features. The multi-sensor temporal feature extraction process is as follows: Based on the extraction of multidimensional features, the variation law of multidimensional features over time is analyzed, and multi-sensor time-series features are extracted, including time-series statistical features, time-series trend features, and time-series abrupt change features.

5. The artificial intelligence-based amorphous alloy motor operation performance monitoring system according to claim 1, characterized in that: The improved LSTM and Attention mechanism obtains historical motor operation data from an existing database and preprocesses it. The preprocessed historical motor operation data is then input into the LSTM network in a time series. The memory unit selectively saves long-term information and captures the long-term dependencies of motor operation data in the time dimension. Furthermore, the Attention mechanism is applied to the output of the LSTM, using the LSTM output as the input of the query, key, and value. The attention weights are obtained by calculating the similarity between the query and the key.

6. The artificial intelligence-based amorphous alloy motor operation performance monitoring system according to claim 5, characterized in that: The fusion Transformer architecture includes encoder construction and positional encoding; The encoder is constructed using the Transformer encoder structure, which includes a multi-head self-attention mechanism and a feedforward neural network. The attention weights obtained by LSTM and the Attention mechanism are input into the Transformer encoder. The multi-head self-attention mechanism maps the input attention weights to different subspaces, generates multiple Query, Key, and Value matrices to capture global dependencies, and concatenates the results of each head for linear transformation. The feedforward neural network enhances the model's learning ability. Positional encoding, typically created using sine and cosine functions, is added to the input attention weights to assign unique information about the position of each attention weight.

7. The artificial intelligence-based amorphous alloy motor operation performance monitoring system according to claim 6, characterized in that: The process for setting the output energy efficiency level is as follows: The feature sequence is first modeled by an improved LSTM to determine the long-term temporal dependence of the parameters, while retaining the dynamic features of energy efficiency assessment. The key variables are then weighted by the Attention mechanism and then fed into the Transformer encoder to capture the global interaction of multiple variables and enhance the discriminative power of the features. By deeply integrating the temporal evolution trend of LSTM with the global correlation of Transformer through the feature fusion layer, a comprehensive representation of the input classification layer is formed, which outputs the probability distribution of IE4 and IE5 energy efficiency levels, and takes the level corresponding to the highest probability as the judgment result.

8. The artificial intelligence-based amorphous alloy motor operation performance monitoring system according to claim 7, characterized in that: The process of predicting the output motor health index and remaining motor lifespan is as follows: We acquire energy efficiency levels and multi-sensor time-series features to form a basic dataset; we construct a graph structure data with multi-sensor nodes and physical field coupling relationships as edges; we globally pool the node features of the last layer of the GNN and input them into a fully connected layer; we comprehensively consider the energy efficiency level decay trend, the anomaly degree of multi-sensor and the stability of coupling relationships, and obtain a health index between 0 and 1 by weighted summation. Using the time series of health index as input, combined with the historical change curve of energy efficiency level, a long short-term memory network is used to fit the health state decay model, learn the evolution law to establish the mapping between the running time, and predict the remaining lifespan by combining the current energy efficiency level decay rate.

9. The artificial intelligence-based amorphous alloy motor operation performance monitoring system according to claim 8, characterized in that: The specific process of the 3D dynamic coupling display of the output motor is as follows: The system integrates motor health index, motor remaining life prediction, and multi-field coupling parameters and environmental conditions of motor operation into a digital twin system. It also identifies the corresponding components of the motor's 3D model with differentiated visualizations and dynamically presents countdown information based on the motor's remaining life prediction. The system integrates simulation algorithms to map the dynamic changes of multi-field coupling parameters and environmental conditions of motor operation onto the model. Finally, it uses graphics rendering and animation technology to fuse information and form a 3D interactive interface.

10. A method for monitoring the operating performance of amorphous alloy motors based on artificial intelligence, characterized in that: Includes the following steps: Step 1: Collect multi-field coupling parameters of motor operation and environmental condition data. The multi-field coupling parameters of motor operation include electrical coupling parameters, thermal coupling parameters and mechanical coupling parameters; Step 2: Preprocess the multi-field coupling parameters of the motor operation and the environmental condition data, and introduce multi-field coupling features to extract multi-dimensional features and multi-sensor time-series features; Step 3: Based on historical motor operation data, an improved LSTM and Attention mechanism is adopted and integrated with the Transformer architecture to build a dynamic mapping model. Multi-dimensional features are input into the dynamic mapping model, and the output is the energy efficiency level and the probability distribution of IE4 and IE5 in the energy efficiency level. The level corresponding to the highest probability is taken as the judgment result. Based on the energy efficiency level, a graph structure data is constructed with multiple sensors as nodes and physical field coupling relationship as edges. The graph neural network is used to fuse the time series features of multiple sensors to output the motor health index and the prediction of the remaining life of the motor. Step 4: Visualize the motor health index and the motor remaining life prediction using digital twins, and output a 3D dynamic coupling display of the motor.

Images