Electromechanical equipment fault diagnosis method and system based on data analysis

Abstract

The invention relates to the technical field of fault diagnosis, in particular to an electromechanical equipment fault diagnosis method and system based on data analysis. The method comprises the following steps: extracting a historical operation log of a motor, analyzing a current overload state and constructing a current overload spectrogram; then, carrying out insulation layer performance loss gradient analysis based on the spectrogram, and calculating the failure probability of the motor; finally, a risk level is given to the overload current according to the failure probability, an equipment fault diagnosis framework is constructed, and the framework is sent to a control terminal to execute fault diagnosis. According to the invention, the electromechanical equipment fault diagnosis technology is optimized, so that the electromechanical equipment fault diagnosis technology is more accurate.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a method and system for diagnosing electromechanical equipment faults based on data analysis. Background Art

[0002] The application of data analysis technology offers a new approach to fault diagnosis in electromechanical equipment. By collecting and analyzing equipment operating data, especially by deeply exploring historical operating data, potential failure modes and trends can be revealed. Modern electromechanical equipment is typically equipped with various sensors that monitor the equipment's operating status in real time, collecting a large amount of operational data, including current, voltage, temperature, vibration, and other information. Multi-dimensional analysis of this data enables comprehensive understanding of the equipment's status. In particular, in fault diagnosis of critical equipment such as motors, data analysis technology can identify abnormalities in motor operation, such as current overload, temperature rise, and abnormal vibration, in real time, and predict the type and timing of potential faults. However, traditional methods for fault diagnosis of electromechanical equipment based on data analysis suffer from inaccurate analysis of insulation loss within motors. This leads to large errors in the calculation of motor failure probability and an inability to accurately diagnose potential motor faults. Summary of the Invention

[0003] Based on this, it is necessary to provide a method and system for diagnosing electromechanical equipment faults based on data analysis to solve at least one of the above technical problems.

[0004] To achieve the above object, a method for diagnosing electromechanical equipment faults based on data analysis is provided, the method comprising the following steps: Step S1: extracting the historical operation log of the motor from the control terminal of the electromechanical equipment; analyzing the current overload status of the historical operation log and constructing a current overload spectrum diagram; Step S2: performing insulation layer performance loss gradient analysis based on the current overload spectrum to obtain the insulation layer performance loss gradient; performing motor failure probability calculation based on the insulation layer performance loss gradient to obtain the motor failure calculation probability; Step S3: Assign an overload current risk level to the current overload spectrum diagram based on the motor failure calculation probability to obtain the overload current risk level; construct an equipment fault diagnosis architecture based on the overload current risk level; and send the equipment fault diagnosis architecture to the control terminal of the electromechanical equipment to perform diagnosis of the electromechanical equipment fault.

[0005] Preferably, step S1 includes the following steps: Step S11: extracting the historical operation log of the motor from the control terminal of the electromechanical equipment; Step S12: extracting the motor overload operation state from the motor's historical operation log to obtain the overload operation state; Step S13: performing current overload state analysis on the overload operation state and constructing a current overload spectrum diagram.

[0006] Preferably, step S2 includes the following steps: Step S21: performing overload multiple evolution analysis based on the current overload spectrum diagram to obtain overload multiple evolution data; Step S22: Calculating the line temperature rise cumulative gain index based on the overload multiple evolution data to obtain the line temperature rise cumulative index; Step S23: performing insulation layer performance loss gradient analysis based on the line temperature rise cumulative index to obtain the insulation layer performance loss gradient; Step S24: Calculating the motor failure probability based on the overload multiple evolution data and the insulation layer performance loss gradient to obtain the motor failure calculation probability.

[0007] Preferably, step S22 includes the following steps: Step S221: performing an incremental multiple change rate analysis on the overload multiple evolution data to obtain an incremental trend multiple change rate; Step S222: performing mutation time series integration on the overload multiple evolution data according to the increasing trend multiple change rate, thereby obtaining the increasing mutation integral multiple; Step S223: performing aggregation of equivalent calorific value gains per unit time according to the incremental mutation integration multiple to obtain a calorific value gain aggregation sequence; Step S224: performing recursive superposition regression analysis on the heat gain series per unit time to generate a line temperature rise accumulation curve; performing exponential fitting processing on the line temperature rise accumulation curve to obtain temperature rise accumulation exponential fitting data; Step S225: Calculating the line temperature rise cumulative gain index based on the temperature rise cumulative index fitting data to obtain the line temperature rise cumulative index.

[0008] Preferably, step S23 includes the following steps: Step S231: Obtain basic properties of line insulation materials; Step S232: constructing a continuous temperature rise variation curve based on the line temperature rise cumulative index, and then calculating the average difference of the temperature rise inflection point increasing rate to obtain the temperature rise inflection point increasing rate difference; Step S233: performing slope growth convergence analysis on the continuous temperature rise change curve according to the temperature rise inflection point growth rate difference to obtain the temperature rise growth convergence range; Step S234: performing a numerical evolution analysis of thermal degradation of the basic characteristics of the line insulation material based on the temperature rise inflection point growth rate difference and the temperature rise growth convergence interval to obtain a numerical evolution law of thermal degradation of the material; Step S235: performing insulation layer performance loss gradient analysis according to the numerical evolution law of material thermal degradation to obtain the insulation layer performance loss gradient.

[0009] Preferably, step S234 includes the following steps: Based on the temperature rise inflection point growth rate difference and the temperature rise growth convergence range, the thermal cycle stress fluctuation simulation analysis of the basic characteristics of the line insulation material is carried out to obtain the material thermal stress fluctuation data; Based on the material thermal stress fluctuation data, the thermal shear plasticity offset of the basic characteristics of the line insulation material is simulated to obtain the thermal shear plasticity offset data; The material plasticizer loss index is quantified based on the material thermal stress fluctuation data and the thermal shear plasticity offset data to obtain the material plasticizer loss index; Based on the thermal shear plasticity offset data and the material plasticizer loss index, the basic characteristics of the line insulation material are numerically analyzed for the thermal degradation evolution to obtain the numerical evolution law of the material thermal degradation.

[0010] Preferably, step S24 includes the following steps: Step S241: performing a high-frequency current oscillation waveform analysis based on the overload multiple evolution data to obtain a high-frequency current oscillation waveform; performing a steep and disordered oscillation analysis on the high-frequency current oscillation waveform to obtain a steep and disordered oscillation waveform; Step S242: performing multi-frequency transient mutation rate coupling on the steep disordered oscillation waveform to obtain multi-frequency transient mutation coupling data; Step S243: Deducing the corresponding failure boundary of the insulation layer based on the insulation layer performance loss gradient according to the multi-frequency transient mutation coupling data to obtain insulation layer failure boundary data; Step S244: performing a winding short circuit multi-factor superposition risk probability calculation based on the multi-frequency transient mutation coupling data and the insulation layer failure boundary data, thereby obtaining the winding short circuit risk probability; Step S245: Calculate the motor failure probability based on the winding short circuit risk probability to obtain the motor failure calculation probability.

[0011] Preferably, step S3 includes the following steps: Step S31: normalizing the calculated probability of motor failure to obtain a normalized probability of motor failure; Step S32: assigning an overload current risk level to the current overload spectrum diagram based on the normalized probability of motor failure to obtain an overload current risk level; Step S33: constructing an equipment fault diagnosis framework based on the normalized probability of motor failure and the overload current risk level; Step S34: Send the device fault diagnosis framework to the control terminal of the electromechanical device to perform diagnosis of the electromechanical device fault.

[0012] Preferably, the present invention further provides a method for diagnosing electromechanical equipment faults based on data analysis, which is used to perform the above-mentioned method for diagnosing electromechanical equipment faults based on data analysis. The electromechanical equipment fault diagnosis system based on data analysis includes: The overload status analysis module is used to extract the historical operation log of the motor from the control terminal of the electromechanical equipment; analyze the current overload status of the historical operation log and construct a current overload spectrum diagram; A failure probability calculation module is used to analyze the insulation layer performance loss gradient based on the current overload spectrum to obtain the insulation layer performance loss gradient; and to calculate the motor failure probability based on the insulation layer performance loss gradient to obtain the motor failure calculation probability; A fault diagnosis architecture building module is used to assign an overload current risk level to the current overload spectrum diagram based on the motor failure calculation probability to obtain the overload current risk level; build an equipment fault diagnosis architecture according to the overload current risk level; and send the equipment fault diagnosis architecture to the control terminal of the electromechanical equipment to perform the diagnosis of the electromechanical equipment fault.

[0013] The present invention has the beneficial effect of extracting historical motor operation logs from the control terminal of an electromechanical device to comprehensively collect motor operating data, particularly current fluctuation information. Current overload status analysis of historical operation logs effectively identifies whether the motor is overloaded, and visualization analysis is performed by constructing a current overload spectrum. This analysis not only facilitates early detection of potential current overload issues but also provides an accurate data foundation for subsequent fault prediction and diagnosis, significantly improving the early warning capabilities of motor fault detection. Analysis of insulation layer performance loss gradients based on the current overload spectrum provides a deep understanding of insulation layer loss over long-term operation. Analysis of insulation layer performance loss gradients accurately assesses the health of the motor's insulation material and predicts whether the motor is at risk of insulation layer damage. Furthermore, calculating the motor failure probability based on the insulation layer loss gradient helps comprehensively understand the motor's reliability, providing a more scientific basis for equipment maintenance decisions and avoiding motor failure or more serious faults caused by insulation layer damage. By assigning an overload current risk level to the current overload spectrum based on the calculated motor failure probability, potential risks in motor operation can be quantified and a more intuitive risk level provided to equipment managers. This risk level analysis can help technicians identify high-risk equipment in a timely manner, develop priority maintenance plans, and reduce the probability of equipment failure. By constructing an equipment fault diagnosis architecture and sending it to the control terminal of the electromechanical equipment, real-time monitoring and automatic diagnosis of equipment failures can be achieved. This closed-loop fault diagnosis system not only improves the accuracy of motor fault detection, but also enhances the efficiency of equipment maintenance, reduces equipment downtime and maintenance costs. Therefore, the present invention is an optimization of a traditional electromechanical equipment fault diagnosis method based on data analysis, which solves the problem that a traditional electromechanical equipment fault diagnosis method based on data analysis does not accurately analyze the loss of the line insulation layer in the electromechanical equipment motor, resulting in large errors in the accuracy of the motor function failure probability calculation and the inability to accurately diagnose the potential faults of the motor. It improves the accuracy of the analysis of the loss of the line insulation layer in the electromechanical equipment motor, reduces the error in the motor function failure probability calculation, and enhances the diagnostic capability of the potential faults of the motor. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 A schematic flow chart of the steps of a method for diagnosing electromechanical equipment faults based on data analysis; Figure 2 for Figure 1 Detailed implementation steps of step S2 in FIG. Figure 3 for Figure 1 Detailed implementation steps of step S3 in FIG. DETAILED DESCRIPTION

[0015] See also Figures 1 to 3 , a method for diagnosing electromechanical equipment faults based on data analysis, the method comprising the following steps: Step S1: extracting the historical operation log of the motor from the control terminal of the electromechanical equipment; analyzing the current overload status of the historical operation log and constructing a current overload spectrum diagram; Step S2: performing insulation layer performance loss gradient analysis based on the current overload spectrum to obtain the insulation layer performance loss gradient; performing motor failure probability calculation based on the insulation layer performance loss gradient to obtain the motor failure calculation probability; Step S3: Assign an overload current risk level to the current overload spectrum diagram based on the motor failure calculation probability to obtain the overload current risk level; construct an equipment fault diagnosis architecture based on the overload current risk level; and send the equipment fault diagnosis architecture to the control terminal of the electromechanical equipment to perform diagnosis of the electromechanical equipment fault.

[0016] In the embodiment of the present invention, reference Figure 1 The above is a schematic flow chart of the steps of a method for diagnosing electromechanical equipment faults based on data analysis according to the present invention. In this example, the method for diagnosing electromechanical equipment faults based on data analysis includes the following steps: Step S1: extracting the historical operation log of the motor from the control terminal of the electromechanical equipment; analyzing the current overload status of the historical operation log and constructing a current overload spectrum diagram; In this embodiment of the present invention, 90 consecutive days of historical motor operation log data were retrieved from electromechanical equipment control terminals deployed on industrial production lines. The log data included timestamps, current sampling values, voltage sampling values, operating status identification codes, and overload alarm records. The log data was segmented into time series, with each segment lasting one hour. Current data within each segment was extracted using a sliding window with a sampling period of 500ms and a window size of 60 sampling points. The current mean within each window was compared with the rated current of the motor (set to 42A) to generate an overload ratio sequence. K-means clustering was then performed on the overload ratio sequence to delineate normal and overload segments. All data windows belonging to the overload segment were extracted, and the current frequency domain features of each window were constructed. The current time series was processed using a fast Fourier transform algorithm, and the first 10 main frequency amplitudes were selected as frequency domain vectors. All frequency domain vectors were concatenated to generate a current overload spectrum. This spectrum was then recorded in matrix form, with each row corresponding to the spectrum data of an overload event.

[0017] Step S2: performing insulation layer performance loss gradient analysis based on the current overload spectrum to obtain the insulation layer performance loss gradient; performing motor failure probability calculation based on the insulation layer performance loss gradient to obtain the motor failure calculation probability; In the embodiment of the present invention, based on the constructed current overload spectrum, a key frequency band of 50Hz to 1000Hz is selected to analyze the insulation layer performance loss gradient. A mathematical model of thermal aging of insulation materials is established, and the calculation method of the insulation layer performance loss gradient G in the model is as follows: , where α is the material characteristic coefficient, T is the temperature value, β is the temperature influence index, f is the frequency value, γ is the frequency influence index, t is the action time, and δ is the time accumulation index. The preset parameter values ​​are obtained through the preset insulation material accelerated aging test records: α is 0.025, β is 1.8, γ is 0.6, and δ is 0.4. The power spectrum density value of each frequency point in the current overload spectrum diagram is input into the model as the frequency influence factor. Combined with the equipment operating temperature data and overload duration data, the insulation layer performance loss gradient value corresponding to each frequency point is calculated. A motor failure probability calculation model is established, and the calculation method of the failure probability P is: ,in is the insulation layer performance loss gradient value at the i-th frequency point, The corresponding weight coefficient is determined through statistical analysis of historical fault data. The weight coefficient is 0.8 for the high-frequency band, 0.6 for the medium-frequency band, and 0.4 for the low-frequency band. The weighted loss gradient values ​​at all frequency points are accumulated and summed, and the motor failure probability is calculated using an exponential function transformation.

[0018] In another embodiment, the current overload spectrograms constructed in step S1 are first arranged in chronological order, and the spectral energy density is extracted for each frame. Specifically, within each spectrogram, the low-frequency band (0-100 Hz), mid-frequency band (100-300 Hz), and high-frequency band (300-1000 Hz) are integrated to obtain the total energy values ​​for the three frequency bands. These three energy values ​​are combined into a spectral energy feature vector, representing the energy distribution characteristics of each overload event in the frequency domain. Next, a time series analysis is performed on these spectral energy feature vectors, and a local weighted regression method is used to fit the energy variation trend over time to identify the pattern of energy accumulation over time. To extract indicators related to thermal effects, the fitted energy variation trend chart is segmented into 5-hour periods. The energy growth slope in each segment is statistically analyzed to obtain a series of energy accumulation rates during the overload process. Subsequently, a line temperature rise model is constructed based on the energy accumulation rate series. A typical heat conduction path is selected as the analysis object, and the cable length is set to 10 meters, the cross-sectional area is 4 mm², and the copper conductor resistivity is The heat input per unit time is calculated using the square of the current multiplied by the resistance and the power-on time. Combined with the thermal conductivity between the conductor and the environment (set to 15 W / m²·K), a heat balance analysis is performed to derive the cumulative line temperature rise per unit time. The energy accumulation rate for each segment is converted into a line temperature rise value, which is then accumulated to form a temperature rise time series. The temperature rise series is further tested for peak growth rates. Local second-order difference analysis is performed on segments with more than three consecutive temperature rises to identify inflection points of sudden temperature rise. The slope difference and occurrence time of each inflection point are recorded to establish a time interval distribution for these inflection points. The line temperature rise cumulative gain index is then calculated based on the average temperature rise increase between these inflection points. This index reflects the degree of coupling between the cumulative increase in line temperature rise per unit time and insulation aging. Next, based on the resulting temperature rise cumulative gain index, an insulation layer performance loss gradient analysis is performed. Based on the insulation material's thermal resistance grade, for example, the allowable long-term operating temperature for Class E insulation is 120°C. The temperature rise cumulative gain index is compared with this temperature limit, and a thermal degradation distribution map is generated for the temperature rise gain intervals exceeding the limit. Combining the material aging rate curve with laboratory aging data, each gain index segment is converted into a loss gradient value. Interpolation is then used to complete the uncovered time periods, forming a complete loss gradient time series. Finally, the motor failure probability is calculated using this insulation layer performance loss gradient time series. A gradient-probability mapping table is constructed based on historical statistics of actual failure rates for similar motors at various loss gradient levels. The average loss gradient value of the current motor over the last 10 hours is substituted into this mapping table to determine the corresponding failure rate. For example, when the average loss gradient is 0.76, historical data indicates that 32% of motors experience insulation breakdown within 100 hours. Combined with the current total operating time of the equipment and known maintenance cycle data, a proportional conversion is performed, resulting in a calculated failure probability of 27%. This value is then passed to the next risk level assignment process and serves as a core input indicator for failure risk assessment.

[0019] Step S3: Assign an overload current risk level to the current overload spectrum diagram based on the motor failure calculation probability to obtain the overload current risk level; construct an equipment fault diagnosis architecture based on the overload current risk level; and send the equipment fault diagnosis architecture to the control terminal of the electromechanical equipment to perform diagnosis of the electromechanical equipment fault.

[0020] In this embodiment of the present invention, the motor failure calculation probability obtained in step S2 is linearly normalized, so that all failure probability values ​​fall within the decimal range between 0 and 1. The normalized failure probability values ​​are mapped to five risk levels, with the following classification rules: 0–0.2 is level 1 risk, 0.2–0.4 is level 2 risk, and so on, up to 0.8–1.0 for level 5 risk. Based on the risk level, the spectrum vector of each time period in the current overload spectrum graph is associated with the current level, and each time period is labeled with a risk label, forming a spectrum graph annotated with the risk level. Subsequently, an equipment fault diagnosis architecture is constructed, which takes three core information types as input: the current spectrum risk level distribution, the loss gradient score sequence, and the failure calculation probability. Nodes are organized in a graph structure, with node attributes including: spectrum time window, risk level, estimated failure time, and recommended maintenance action. A hierarchical priority parsing algorithm is used to generate a fault path sequence, outputting the associated risk propagation paths and maintenance priority order between nodes. Finally, the diagnostic architecture is packaged in JSON format and sent to the monitoring system of the electromechanical equipment control terminal through the industrial Ethernet interface. It is then embedded in the control logic trigger module to facilitate real-time execution of the motor fault diagnosis and risk warning process.

[0021] Step S1 includes the following steps: Step S11: extracting the historical operation log of the motor from the control terminal of the electromechanical equipment; Step S12: extracting the motor overload operation state from the motor's historical operation log to obtain the overload operation state; Step S13: performing current overload state analysis on the overload operation state and constructing a current overload spectrum diagram.

[0022] In an embodiment of the present invention, a 60-day historical operation log of the motor is extracted from a PLC control system. The log content includes current values, voltage values, operation time stamps, operation status codes (status code 2000 indicates "normal operation" and 2100 indicates "overload operation"), power factor, and thermal protection relay tripping records, recorded every 5 seconds. A total of more than one million operation records are extracted. During the extraction process, the log data is sorted according to the recording time field, records with duplicate and missing timestamps are removed, and outliers (such as records with current values ​​greater than 300A) are corrected to maintain data accuracy. The extracted log is stored as a CSV format file and loaded into a data processing module for subsequent analysis. The overload operation status is extracted from the extracted motor historical operation log. Using the rated current of 42A as the reference baseline, all recorded current data is processed with an analysis cycle of 5 seconds. The overload judgment threshold is set to 1.1 times the rated current, that is, 46.2A. When the current of three consecutive analysis cycles is greater than this threshold and the status code is displayed as "operating", the interval is marked as a "stable overload section". The boundary of the stable overload section is further expanded, extending 2 cycles forward and backward respectively, to extract the complete overload operating state. For each overload section, all current sequences and voltage sequences in the corresponding original time period are extracted, and the average current value, peak current, root mean square current, peak duration, power factor variation range and tripping event correlation of each section are calculated. A total of 372 valid overload sections are extracted as analysis samples. Each overload sample is saved as an independent data record with a complete timestamp, operating status information and corresponding physical quantity parameters.

[0023] The extracted overload operating states were analyzed for current overload conditions, and a current overload spectrum was constructed. Data segments longer than 60 seconds within each overload segment were used as target data. Current sequences were extracted and preprocessed, including linear interpolation to correct missing values, bandpass filtering to remove interference from the 50Hz grid fundamental frequency (using a 0.1Hz to 40Hz bandpass filter), and amplitude normalization for each segment. A fast Fourier transform (FFT) was then performed on each current segment, using a 1024-point window and a Hanning window function. Frequency amplitude data within the 0-500Hz range were extracted from the resulting spectrum data, yielding 500 frequency amplitude points for each sample. For uniform representation, the 500-dimensional frequency vectors were unified into a two-dimensional spectrum matrix, with rows representing different overload event samples and columns representing the amplitudes at the corresponding frequency points. To facilitate subsequent analysis, the total energy of the low-frequency band (0-100Hz), mid-frequency band (100-300Hz), and high-frequency band (300-500Hz) was further extracted to construct a three-dimensional energy feature vector, which served as the spectral summary for each overload event. Furthermore, to identify the temporal evolution characteristics of spectral changes, the spectrum samples were arranged in chronological order to form a spectrum evolution sequence, which supported subsequent in-depth analysis of insulation performance changes and failure probabilities. Finally, the constructed current overload spectrum was stored in a matrix structure and linked to the time stamp of each overload segment in the original log using a primary key, enabling synchronous data access and analysis.

[0024] Step S2 includes the following steps: Step S21: performing overload multiple evolution analysis based on the current overload spectrum diagram to obtain overload multiple evolution data; Step S22: Calculating the line temperature rise cumulative gain index based on the overload multiple evolution data to obtain the line temperature rise cumulative index; Step S23: performing insulation layer performance loss gradient analysis based on the line temperature rise cumulative index to obtain the insulation layer performance loss gradient; Step S24: Calculating the motor failure probability based on the overload multiple evolution data and the insulation layer performance loss gradient to obtain the motor failure calculation probability.

[0025] As an example of the present invention, refer to Figure 2 As shown, in this example, step S2 includes: Step S21: performing overload multiple evolution analysis based on the current overload spectrum diagram to obtain overload multiple evolution data; In this embodiment of the present invention, the spectrum of each overload sample constructed in step S1 is synchronously processed with the raw current value corresponding to that sample, and the ratio of the maximum effective current value to the motor's rated current value is extracted to obtain the overload multiple value. The motor's rated current is set to 42A, and the peak current value in each sample is divided by 42 to form an overload multiple sequence. All overload events are then arranged in chronological order, and the difference between two adjacent overload multiples is calculated to form an overload multiple difference sequence. This sequence represents the changing trend of overload intensity over consecutive events. To describe its evolution, a sliding average with a fixed window width of 10 events is used, and a differential technique is used to extract the local rate of change. The results of the third-order differential analysis are then used to construct an evolution trend graph. Each data point in the evolution trend graph represents the upward or downward trend of the overload intensity within the current time window. The positive or negative direction indicates the direction of the trend change, and the absolute value indicates the rate of change. Finally, the evolution trend graph is output as the overload multiple evolution data, and the location of the evolution inflection point is recorded for dynamic segmentation processing in subsequent temperature rise analysis.

[0026] Step S22: Calculating the line temperature rise cumulative gain index based on the overload multiple evolution data to obtain the line temperature rise cumulative index; In the embodiment of the present invention, the line temperature rise cumulative gain index is calculated based on the overload multiple evolution data obtained in step S21. First, the duration of each overload event and its corresponding overload multiple value are extracted and regarded as the heat input parameter per unit time. The calculations are based on a conductor cross-sectional area of ​​4 mm² and a length of 10 m. According to Joule's law, the square of the overload current is multiplied by the resistance and then by the duration to calculate the total heat input for each event. The heat input values ​​of all events are arranged in chronological order and cumulatively summed to construct a temperature rise accumulation curve. To characterize the dynamic gain during the temperature rise accumulation process, the cumulative growth rate is calculated for each unit interval of 10 events and compared with the previous interval to form a gain ratio sequence. A gain ratio greater than 1.2 is used as the threshold to screen out growth mutation points. Based on the time interval and growth amplitude between each mutation point, a weighted average method is used to form a temperature rise cumulative gain index. This index is used to describe the increasing trend of temperature rise over time during the line overload process. Higher values ​​indicate more concentrated and intense temperature rise growth.

[0027] In another embodiment, a line temperature rise calculation method is established based on the acquired overload factor evolution data. The temperature rise value is equal to the square of the overload factor multiplied by a baseline temperature rise value, multiplied by a time accumulation coefficient, where the baseline temperature rise value is set to 15°C. The time accumulation coefficient is expressed using an exponential decay function as a natural constant raised to the power of -0.1 times the time interval. A cumulative gain index calculation process is established. First, the overload factor evolution data is grouped into 5-minute intervals. The maximum overload factor within each group is taken as the representative value for that time period. A recursive accumulation algorithm is used to process the temperature rise data for each time period. The cumulative temperature rise for the current time period is equal to the cumulative temperature rise for the previous time period multiplied by the attenuation coefficient plus the additional temperature rise for the current time period. The attenuation coefficient is set to 0.95. A gain index quantification method is established. The gain index is equal to the cumulative temperature rise value divided by the baseline cumulative temperature rise value. The baseline cumulative temperature rise value is determined to be 180°C based on a temperature rise test of 24 hours of continuous operation under rated conditions. The calculated gain index is smoothed using a sliding average algorithm. The sliding window length is set to 12 data points, corresponding to a 1-hour time span. A gain index correction mechanism is established to account for the influence of ambient temperature. When the ambient temperature exceeds 25°C, the gain index is multiplied by a temperature correction factor equal to 1 plus the ambient temperature minus 25°C, divided by 100. The final value of the line temperature rise cumulative index is obtained by taking a weighted average of the gain indices for all time periods. The weight factor is inversely proportional to the time distance, with the weight for the most recent time period being 1.0 and decreasing by 0.1 for each hour forward.

[0028] Step S23: performing insulation layer performance loss gradient analysis based on the line temperature rise cumulative index to obtain the insulation layer performance loss gradient; In this embodiment of the present invention, a sliding window of 30-minute units is selected from the data sequence of the line temperature rise cumulative index obtained in step S22. The growth rate of the temperature rise cumulative index is extracted within each sliding window, and its incremental trend sequence is obtained using the first-order backward difference method. Subsequently, a convolution smoothing kernel function is used to reduce noise in the incremental trend sequence to eliminate the interference of non-periodic fluctuations on the insulation performance analysis. Based on this, an insulation material aging response function model is introduced for each temperature rise incremental window sequence. The input of this model is the rate of change of the line temperature rise, and the output is the ratio of the dielectric strength decrease per unit time. This response function is constructed based on the mathematical expression for thermal aging and life curves in the IEC 60216 standard. A nonlinear least squares fitting method is used to fit the historical temperature rise response experimental data, ultimately obtaining a thermal stress-insulation performance response curve suitable for the current motor model. During the calculation, each temperature rise increment at the current moment is matched with its corresponding response function to obtain the theoretical insulation performance degradation amplitude within that time window. By superimposing the degradation amplitudes within each time window in a time series, a time-varying insulation performance loss rate sequence is obtained. Furthermore, to capture the gradient distribution of insulation performance loss, a first-order derivative analysis of the loss rate sequence is performed to obtain insulation performance loss gradient data. This gradient reflects the dynamic trend of insulation performance degradation over time and serves as an important input variable for the failure probability calculation in the subsequent step S24.

[0029] In another embodiment, an insulation layer performance loss analysis algorithm was established based on line temperature rise cumulative index data, using Arrhenius thermal aging theory as the analysis basis. The insulation material was set to polyimide film, 0.025 mm thick, with a temperature rating of 180°C. After 20,000 hours of continuous operation at this temperature, the insulation strength dropped to 50% of its initial value. A temperature-life relationship data table was established, with the insulation life halved for every 10°C increase. A table lookup method was used to obtain the insulation performance loss ratio corresponding to different temperature rise cumulative indices. A gradient descent algorithm was used to analyze the changing trend of insulation layer performance loss, with a learning rate of 0.01, 1000 iterations, and a convergence criterion of less than 0.0001 between two consecutive iterations. A multilayer perceptron neural network structure was established for insulation performance prediction. The network consisted of an input layer, two hidden layers, and an output layer. The input layer had five nodes corresponding to parameters such as the temperature rise cumulative index, time, voltage, humidity, and frequency. The first hidden layer had ten nodes, the second hidden layer had eight nodes, and the output layer had one node corresponding to the insulation performance loss gradient. The network was trained using the backpropagation algorithm, with the hyperbolic tangent function as the activation function and the mean square error as the loss function. The training dataset consisted of 5,000 sets of historical insulation performance test data. During gradient analysis, the finite difference method was used to calculate the spatial and temporal gradients of insulation performance loss, with a spatial step size of 0.001 mm and a temporal step size of 1 hour.

[0030] Step S24: Calculating the motor failure probability based on the overload multiple evolution data and the insulation layer performance loss gradient to obtain the motor failure calculation probability.

[0031] In this embodiment of the present invention, the calculation of motor failure probability based on overload multiple evolution data and insulation layer performance loss gradients requires first discretizing the overload multiple evolution data obtained in step S21 into a time series. The overload multiples within the entire monitoring period are discretely labeled at fixed time intervals (e.g., 1-second intervals), and an overload time series matrix is ​​constructed. The overload multiple evolution data is represented as a multidimensional vector sequence, where the vector element at each moment represents the overload multiple, instantaneous overload amplitude change rate, and cumulative frequency of occurrence recorded at that moment. Next, the insulation layer performance loss gradient obtained in step S23 is aligned with the overload multiple time series according to the corresponding timestamp to construct a joint analysis matrix. To construct a motor failure probability assessment system, conditional probability theory in statistics is used for modeling, with the overload multiple state and insulation layer loss gradient state used as a priori factors for the occurrence of a failure event. Motor failure is set as the target event, and the frequency of motor failures at different overload multiple intervals and different loss gradient levels is calculated based on a historical motor failure database. Using frequency statistics, an empirical probability distribution table is established, and conditional probabilities are updated according to Bayes' theorem to construct a joint-state conditional probability model for motor failure. Specifically, let the overload multiple state be A, the insulation layer loss gradient state be B, and the motor failure event be C. The probability of failure event C occurring in both states A and B is calculated. To further improve the accuracy of failure probability inference, the K-nearest neighbor (K-NN) algorithm is introduced to cluster and classify historical sample data. Based on similar sample groups in the feature space of the current joint states A and B, the proportion of failure events is extracted as an approximate estimate of the motor failure probability in the current state. The K value is set to 10, and the similarity between joint state vectors is measured based on Euclidean distance, with closer samples receiving higher weight in the final result. Finally, all estimated results are weighted averaged to obtain the calculated motor failure probability for the target state, outputting a real probability value between 0 and 1.

[0032] In another embodiment, based on overload multiple evolution data, a bandpass filter is used to extract high-frequency current oscillation components. The filter passband range is set to 500Hz to 5000Hz, the filter order is an 8th-order Butterworth filter, the passband ripple is less than 0.5dB, and the stopband attenuation is greater than 40dB. The filtered high-frequency signal is subjected to a Hilbert transform to obtain an analytical signal, and the instantaneous amplitude and frequency information are extracted. Signal segments with an instantaneous amplitude change rate exceeding 10% per second and an instantaneous frequency offset exceeding 5% of the fundamental frequency are marked as steep, disordered oscillation waveforms. A wavelet transform is used to analyze the multi-scale characteristics of steep oscillation waveforms, using the Daubechies-8 wavelet basis function and a decomposition level of 6 to obtain wavelet coefficients for 6 frequency bands. Transient mutation detection is performed on the wavelet coefficients of each frequency band, and an adaptive threshold method is used to identify mutation points. The threshold is set to 3 times the standard deviation of the wavelet coefficients. A transient mutation event is identified when the wavelet coefficient amplitude exceeds the threshold and the duration is less than 0.1s. A multi-frequency transient mutation rate coupling matrix was established. The matrix dimension was 6×6, corresponding to the coupling relationships between the six frequency bands. The matrix elements were the time correlation coefficients of the mutation events between the corresponding frequency bands. The correlation coefficients were obtained using Pearson correlation analysis and ranged from -1 to 1. A failure boundary criterion was established based on the insulation layer performance loss gradient data. When the insulation layer performance loss gradient exceeded 0.5% per hour and the multi-frequency transient mutation coupling strength was greater than 0.7, the insulation layer reached the failure boundary. A Bayesian inference method was used to calculate the winding short-circuit risk probability. The prior probability was set to the historical winding short-circuit fault frequency of 0.02%. A likelihood function was constructed based on the multi-frequency transient mutation coupling data and the insulation layer failure boundary data. The likelihood function value was equal to the coupling strength multiplied by the insulation loss gradient, divided by the normalization constant 100. The posterior probability (i.e., the winding short-circuit risk probability) was equal to the prior probability multiplied by the likelihood function value, divided by the marginal probability, which was calculated using the total probability formula. The final motor failure calculation probability is obtained using the fault tree analysis method. A fault tree structure is established that includes failure modes such as winding short circuit, bearing fault, and insulation aging. The failure modes are connected using OR gate logic. The motor failure probability is equal to 1 minus the product of the reliabilities of all failure modes. The winding short circuit reliability is equal to 1 minus the winding short circuit risk probability. The bearing failure reliability is determined to be 0.98 based on the bearing life distribution function, and the insulation aging reliability is determined based on the insulation performance loss gradient.

[0033] Step S22 includes the following steps: Step S221: performing an incremental multiple change rate analysis on the overload multiple evolution data to obtain an incremental trend multiple change rate; Step S222: performing mutation time series integration on the overload multiple evolution data according to the increasing trend multiple change rate, thereby obtaining the increasing mutation integral multiple; Step S223: performing aggregation of equivalent calorific value gains per unit time according to the incremental mutation integration multiple to obtain a calorific value gain aggregation sequence; Step S224: performing recursive superposition regression analysis on the heat gain series per unit time to generate a line temperature rise accumulation curve; performing exponential fitting processing on the line temperature rise accumulation curve to obtain temperature rise accumulation exponential fitting data; Step S225: Calculating the line temperature rise cumulative gain index based on the temperature rise cumulative index fitting data to obtain the line temperature rise cumulative index.

[0034] In this embodiment of the present invention, a numerical differentiation algorithm is established for the acquired overload multiple evolution data, and a five-point central difference method is used for rate of change analysis. The five-point central difference formula is: the rate of change at the current point is equal to the sum of the two previous data points plus 8 times the previous data point minus 8 times the next data point plus the next two data points, divided by 12 times the time interval. The time interval is set to 1 minute, the overload multiple data accuracy is 0.01 times, and the rate of change is expressed in times per minute, with an accuracy of 0.001 times per minute. An increasing trend identification criterion is established: when the rate of change of five consecutive data points is positive and the rate of change value increases, it is marked as an increasing trend segment. The starting and ending points of the increasing trend segment are recorded with a timestamp and the corresponding overload multiple value, respectively. The increasing trend segment is linearly fitted using the sliding least squares method, with the fitting window length set to 10 data points. The slope and intercept of the fitted line are solved using the least squares method. The slope is the increasing multiple rate of change of the trend segment. A rate of change outlier detection mechanism was established, using the 3σ criterion to identify abnormal rates of change. Rates exceeding the mean plus three standard deviations or falling below the mean minus three standard deviations were flagged as outliers and removed. After removing outliers, the increasing trend rate of change data was smoothed using a Gaussian filter. The filter kernel standard deviation was set to 0.5, and the filter window length was seven data points. A smoothed rate of change series was obtained through convolution. A grading standard for increasing trends was established: rates of change between 0.001 and 0.01 times / minute were labeled as slow increases, 0.01 to 0.05 times / minute as moderate increases, and those exceeding 0.05 times / minute as rapid increases. Rate of change data was stored in a time series format, containing fields such as timestamp, raw rate of change, smoothed rate of change, and trend level. Data was sampled once per minute, stored for seven consecutive days, and contained a total of 10,080 data points.

[0035] Based on the increasing trend multiple change rate data, a mutation detection algorithm was established, and the CUSUM cumulative sum control chart method was used to identify the mutation points of the overload multiple evolution. The decision interval of the CUSUM control chart was set to 0.02 times / minute, and the reference value was set to the mean value of the increasing trend change rate of 0.015 times / minute. When the CUSUM value exceeded 5 times the decision interval, that is, 0.1 times / minute, an upward mutation was determined to have occurred. A mutation time series integration algorithm was established, and a trapezoidal integration operation was performed on each identified mutation segment. The integration interval was from the start time of the mutation to the end time of the mutation, and the integration function was a function of the increasing trend multiple change rate with respect to time. In the trapezoidal integration method, the integration interval was divided into n subintervals, and the subinterval length was set to 1 minute. The trapezoidal area formula was used for numerical integration in each subinterval. The trapezoidal area was equal to the subinterval length multiplied by the average value of the function value at the two end points of the interval. An integration multiplier accumulation mechanism was established to superimpose multiple sudden integral values ​​within the same time window. The time window length was set to 30 minutes, with a sliding step of 5 minutes. The accumulated integration multiplier within the window was equal to the algebraic sum of all sudden integral values. An exponentially weighted moving average algorithm was used to enhance the trend of the integration multiplier. The smoothing coefficient was set to 0.2. The current smoothed integration multiplier was equal to the smoothing coefficient multiplied by the current raw integration multiplier plus 1 minus the smoothing coefficient multiplied by the previous smoothed integration multiplier. An incremental sudden integral multiplier normalization process was established to map the integration multiplier value to the range of 0 to 1. The normalization formula was the current integration multiplier minus the minimum integration multiplier, divided by the maximum integration multiplier minus the minimum integration multiplier. The maximum and minimum values ​​were obtained from historical data statistics, with the maximum being 0.85 and the minimum being 0.02. The integration multiplier data storage format includes fields such as timestamp, raw integration value, smoothed integration value, and normalized integration value. Data precision was maintained to four decimal places, and the storage period was every 5 minutes.

[0036] Based on the incremental mutation integration multiple data, an equivalent heat generation conversion algorithm was established. The Joule heating effect principle was used to quantify the heat generated by current overload. The baseline heating power corresponding to a unit overload multiple was set to 100 W. The heat generation was equal to the heat generation power multiplied by the action time, with the time unit being minutes and the heat generation unit being W / min. A heat generation gain function was established, where the heat generation gain was equal to the square of the incremental mutation integration multiple multiplied by the baseline heating power multiplied by the time increment, with the time increment set to a 5-minute data sampling interval. The heat generation distribution within the conductor was analyzed using the heat conduction finite difference method. The conductor cross-section was divided into a 10×10 grid with each grid cell measuring 0.5 mm×0.5 mm. A two-dimensional heat conduction difference equation system was established with a constant temperature boundary and an ambient temperature of 25°C as the boundary conditions. A heat generation gain aggregation algorithm was developed to spatially and temporally integrate the heat generation gains of multiple grid cells per unit time. The spatial integration was performed using the Simpson integration method, and the temporal integration was performed using the Runge-Kutta fourth-order method. The aggregation time window was set to 1 hour, and the aggregated heat gain within the window was equal to the time integral of the heat gain at all time points, with an integration step of 1 minute. A heat gain sequence data structure was established, with sequence elements containing attributes such as timestamp, instantaneous heat gain, cumulative heat gain, and average heat gain. The sequence length was 168 elements, corresponding to a one-week time span, and each element represented a 1-hour aggregation result. A moving window variance analysis method was used to assess the volatility of the heat gain sequence. The window length was set to 24 elements, corresponding to 24 hours, and the variance threshold was set to 20% of the average heat gain. Time periods exceeding the threshold were marked as high-volatility periods. The heat gain aggregate sequence data was stored using a linked list structure, with node data types being double-precision floating-point numbers with a precision of 2 decimal places. The sequence was updated once an hour.

[0037] A recursive overlay regression analysis algorithm was developed based on clustered heat gain data, using a Kalman filter recursive approach to account for the cumulative effects of time series data. The state transition matrix was set to a 2×2 identity matrix, the observation matrix to a 1×2 matrix, the diagonal elements of the process noise covariance matrix to 0.01, and the observation noise variance to 0.1. During the recursive overlay process, the cumulative temperature rise at the current moment was equal to the cumulative temperature rise at the previous moment multiplied by the thermal attenuation coefficient plus the current heat gain multiplied by the thermal conductivity coefficient. The thermal attenuation coefficient was set to 0.98, and the thermal conductivity coefficient was set to 0.15°C / (W·min). A linear regression analysis framework was established, and the least squares method was used to trend the recursive overlay data. The regression equation was a linear function, and the regression coefficients and intercept term were solved through matrix operations. The regression analysis window length was set to 48 data points corresponding to 48 hours, with a window sliding step of 1 hour. Each regression analysis output statistical parameters such as the slope, intercept, correlation coefficient, and residual sum of squares. The temperature rise accumulation curve is generated using a cubic spline interpolation algorithm. The interpolation nodes are the output points of the regression analysis. The interpolation interval is 168 consecutive hours, the interpolation step is 0.1 hours, and 1680 temperature rise accumulation data points are generated. The generated line temperature rise accumulation curve is subjected to exponential fitting processing, and the nonlinear least squares Levenberg-Marquardt algorithm is used to solve the exponential function parameters. The exponential function form is that the temperature rise value is equal to the initial temperature rise value multiplied by the time constant of the natural constant multiplied by the power of time. The initial temperature rise value is set to 25 degrees Celsius. The algorithm iteration parameters are set to the maximum number of iterations 500 times, and the convergence tolerance is 0. The damping parameter has an initial value of 0.01, and the parameter update step size is 0.1. During the fitting process, the objective function is established as the sum of the squares of the differences between the observed and fitted values. The gradient information is calculated using the Jacobian matrix, and the Gauss-Newton method is used to update the parameter vector. The quality of the fit is assessed using the coefficient of determination (R-squared), which is required to be greater than 0.95 and the standard error of fit to be less than 2°C. The storage format for the temperature rise cumulative exponential fit data includes fields such as time series, fitted temperature rise value, fitting parameters, and error information. Data accuracy is maintained to three decimal places, with a time resolution of 6-minute intervals, covering 30 consecutive days of historical data.

[0038] Based on the fitted data of the temperature rise accumulation index, an algorithm for calculating the line temperature rise accumulation gain index was established. The calculation formula is: the gain index equals the current fitted temperature rise value minus the baseline temperature rise value, divided by the baseline temperature rise value multiplied by 100%. The baseline temperature rise value is set to a steady-state temperature rise of 40°C under rated operating conditions. The rate of change of the temperature rise accumulation is calculated using a numerical differentiation method using a second-order central difference format. The rate of change formula is: the rate of change at the current point equals the rate of change at the next data point minus the previous data point, divided by twice the time interval, which is set to 6 minutes. A time-weighted gain index algorithm was established, assigning different weights to the gain index in different time periods: a weight of 1.0 for the last 24 minutes, 0.8 for 24 to 72 hours, 0.6 for 72 to 168 hours, and 0.4 for periods beyond 168 hours. The weighted gain index is calculated by multiplying the gain index for each time period by the cumulative sum of the corresponding weight coefficients, divided by the cumulative sum of the weight coefficients. A gain index smoothing algorithm was established, using double exponential smoothing to eliminate data fluctuations. The first smoothing coefficient was set to 0.3, and the second smoothing coefficient was set to 0.4. The gain index after double exponential smoothing was calculated as 2 times the first smoothing value minus the second smoothing value. A gain index anomaly detection mechanism was established, using a control chart method based on statistical process control. The upper control limit was set to the average gain index plus three times the standard deviation, and the lower control limit was set to the average gain index minus three times the standard deviation. Data points exceeding the control limits were marked as outliers and corrected. The correction method used a median filter algorithm with a filter window length of five data points. Outliers were replaced with the median value within the window. The final line temperature rise cumulative index was obtained through polynomial fitting. A cubic polynomial was used to fit the time-varying pattern of the gain index. The fitting coefficient was solved using the Vandermonde matrix. The function value of the fitting polynomial at the current moment was the line temperature rise cumulative index. Cumulative index data validation uses a cross-validation method, dividing historical data into a training set (70%) and a test set (30%). Validation metrics include root mean square error (RMS) and mean absolute error (MAE), with the RMS error required to be less than 5% and the MAE less than 3%. The line temperature rise cumulative index ranges from 0% to 500%, with accuracy maintained to one decimal place.

[0039] Step S23 includes the following steps: Step S231: Obtain basic properties of line insulation materials; Step S232: constructing a continuous temperature rise variation curve based on the line temperature rise cumulative index, and then calculating the average difference of the temperature rise inflection point increasing rate to obtain the temperature rise inflection point increasing rate difference; Step S233: performing slope growth convergence analysis on the continuous temperature rise change curve according to the temperature rise inflection point growth rate difference to obtain the temperature rise growth convergence range; Step S234: performing a numerical evolution analysis of thermal degradation of the basic characteristics of the line insulation material based on the temperature rise inflection point growth rate difference and the temperature rise growth convergence interval to obtain a numerical evolution law of thermal degradation of the material; Step S235: performing insulation layer performance loss gradient analysis according to the numerical evolution law of material thermal degradation to obtain the insulation layer performance loss gradient.

[0040] In the embodiment of the present invention, the basic characteristic parameters of the insulation material are first extracted from the motor technical specifications. The insulation material is a polyimide film with a grade of 6051 and a thickness of 0.025 mm. The basic material properties include thermal conductivity of 0.12 W / (m·K), specific heat capacity of 1.09 kJ / (kg·K), density of 1420 kg / m³, and linear thermal expansion coefficient. The molecular structure characteristic peaks of the insulating material were measured by infrared spectrometer, and the main characteristic peaks were located at Corresponding to the imide group, For the aromatic ring skeleton, the test conditions were room temperature (25°C) and relative humidity (45%). Differential scanning calorimetry was used to determine the material's glass transition temperature (GTP) to be 385°C, and its thermal decomposition temperature to be 580°C. The scanning rate was 10°C / min, nitrogen atmosphere was used, and the flow rate was 50 mL / min. The electrical properties of the insulating material were measured using a dielectric strength tester, with a breakdown voltage of 220 kV / mm, a dielectric constant of 3.5, and a dielectric loss tangent of 0.003. The test frequency was 1 kHz, and the temperature was 23°C. A material property database table structure was established, containing fields such as material number, physical parameters, thermal parameters, electrical parameters, and mechanical parameters. Data accuracy was maintained to four significant digits.

[0041] Based on the line temperature rise accumulation index data, a continuous temperature rise curve was constructed using a cubic spline interpolation algorithm. The interpolation nodes were discrete data points of the temperature rise accumulation index, with a time interval of 1 hour. After interpolation, the temporal resolution was increased to 6-minute intervals. Natural boundary conditions were used for the cubic spline interpolation, namely, the second-order derivatives at both ends of the curve were zero. The spline coefficients were obtained by solving the tridiagonal matrix equations using the pursuit method. A temperature rise inflection point detection algorithm was established, using the second-order derivative sign change method to identify curve inflection points. The second-order derivatives were obtained through numerical differentiation using the central difference method with a step size of 6 minutes. The inflection point determination criteria were that the absolute value of the second-order derivative was greater than 0.01°C / h² and that the signs of the second-order derivatives at adjacent points were opposite. The incremental growth rate of the identified temperature rise inflection points was calculated using the temperature rise value at the current inflection point minus the temperature rise value at the previous inflection point divided by the time interval. The time interval was measured in hours, and the rate was measured in °C / h. Establish an increasing rate sequence data structure. Sequence elements contain attributes such as the inflection point timestamp, inflection point temperature rise value, increasing rate value, and inflection point type. Inflection point types are categorized as maximum, minimum, and inflection points. The arithmetic mean method is used to calculate the mean of the increasing rate at the temperature rise inflection point. The mean is calculated by dividing the sum of all increasing rate values ​​by the number of increasing rate data points. Establish an increasing rate variance analysis. The variance is calculated by subtracting the square of the mean from the increasing rate value divided by the number of data points minus 1. The standard deviation is the square root of the variance. The difference in the increasing rate of the temperature rise inflection point is calculated using the adjacent inflection point rate difference method. The difference in increasing rate is calculated by subtracting the increasing rate of the previous inflection point from the increasing rate of the current inflection point. The value range for the difference in increasing rate is -10°C / h to +10°C / h, with an accuracy of two decimal places. Establish an increasing rate difference statistical analysis, including statistics such as the maximum increasing rate difference, the minimum increasing rate difference, the average increasing rate difference, and the standard deviation of the increasing rate difference. The statistical window is set to 24 hours, with a sliding step size of 1 hour. Growth rate difference data is stored in a time series database format, with the index field being the timestamp, and the data field containing the growth rate difference, statistical parameters, and quality indicators. The data compression ratio is set to 80%, and the storage period is 30 consecutive days. Based on the growth rate difference data at the temperature rise inflection point, the slope of the continuous temperature rise curve is analyzed. The slope is calculated using a moving window linear regression method. The regression window length is set to 12 data points corresponding to a 2-hour time span, and the window sliding step is 1 data point corresponding to a 6-minute interval. The linear regression method uses the least squares method to solve the slope parameter. The regression equation is the temperature rise value equal to the slope multiplied by the time plus the intercept. The unit of the slope is °C / h. A slope growth criterion is established: when the slope values ​​of five consecutive regression windows show an increasing trend and the growth rate is greater than 0.1°C / h per window, it is marked as a slope growth segment. A convergence analysis algorithm is used to test the convergence characteristics of the slope growth. Convergence is determined using the Cauchy convergence criterion: the slope growth segment is considered to have converged when the maximum difference of 10 consecutive slope values ​​is less than 0.05°C / h.A method for determining the convergence interval was established. The convergence interval started at the moment the slope began to stabilize, and ended at the moment the slope rate of change fell below a threshold set at 0.01°C / h². A sliding variance analysis method was used to quantify slope stability, with a variance window length of 20 data points and a variance threshold set at 0.25°C / h². Intervals below the threshold were marked as convergence intervals. A multi-scale convergence analysis framework was established, with three analysis scales: short-term (2 hours), medium-term (8 hours), and long-term (24 hours). Different convergence criteria were used for each scale: a short-term convergence tolerance of 0.02°C / h, a medium-term convergence tolerance of 0.05°C / h, and a long-term convergence tolerance of 0.1°C / h. Convergence interval data was stored, including the interval start time, end time, interval length, convergence slope mean, and convergence variance. The interval length was expressed in hours, with a precision of one decimal place. Convergence interval quality assessment metrics were established, encompassing dimensions such as convergence stability, convergence speed, and convergence persistence. Stability was obtained through variance normalization, speed was expressed as the inverse of the convergence time, and persistence was calculated as the ratio of the convergence interval length to the total analysis time. Temperature rise convergence interval data was stored in an interval tree data structure, supporting efficient interval queries and overlap detection. Tree nodes contain attributes such as interval endpoints, convergence parameters, and subtree pointers. The tree balance factor was maintained above 0.75.

[0042] Based on the temperature rise inflection point growth rate difference and temperature rise growth convergence interval data, a numerical analysis algorithm for material thermal degradation was established, and the Arrhenius equation was used to describe the relationship between temperature and material degradation rate. The activation energy was set to 150kJ / mol, the gas constant was 8.314J / (mol·K), and the reference temperature was 423K corresponding to 150℃. The degradation rate constant was determined by regression of experimental data to be A multi-physics coupling analysis framework was established to consider the synergistic effects of thermal stress, electrical stress, and mechanical stress on the degradation of insulating materials. The thermal stress calculation adopted the linear elastic theory, and the stress value was equal to the elastic modulus multiplied by the thermal expansion coefficient and then multiplied by the temperature difference. The elastic modulus was set to 2.5 GPa, and the thermal expansion coefficient was set to . The finite element method is used to establish the material degradation simulation mesh. The mesh type is tetrahedral element, the element size is 0.1mm, the total number of mesh nodes is 50,000, the total number of elements is 280,000, and the boundary conditions are set as adiabatic boundaries and fixed constraint boundaries. A set of degradation dynamics equations is established, which includes a coupled system of diffusion equation, reaction equation, and heat transfer equation. The backward Euler implicit format is used for time discretization, the time step is set to 0.1h, and the weighted residual method is used for spatial discretization. The Monte Carlo method is used to deal with the randomness of material parameters in the degradation evolution analysis. The random variables include physical parameters such as thermal conductivity, specific heat capacity, and density. The parameter distribution type is normal distribution, the coefficient of variation is set to 10%, and the number of simulations is 1,000. A degradation index quantification system is established. The main indicators include molecular chain breakage rate, crystallinity change rate, dielectric constant drift rate, etc. The molecular chain breakage rate is calculated by the change in infrared spectrum absorption peak intensity. The benchmark intensity is 1776 The fracture rate is measured at the current strength minus the initial strength, divided by the initial strength multiplied by 100%. The numerical evolution of material thermal degradation is represented using a polynomial function fit with a fifth-order fit. The fitting coefficients are calculated using the least squares method. Goodness of fit requires an R² value greater than 0.95 and a residual standard error less than 5%. Evolution data is stored in a time series database with a time resolution of 1 hour. Parameter fields include degradation rate, degradation velocity, and cumulative degradation.

[0043] Based on the numerical evolution data of material thermal degradation, an algorithm for analyzing the performance loss gradient of the insulation layer was established. The finite difference method was used to calculate the gradient distribution of performance loss in both spatial and temporal dimensions. Spatial gradients were calculated using a central difference scheme, where the gradient value is equal to the difference between the performance loss values ​​at adjacent grid points divided by the grid spacing, which was set to 0.05 mm. Forward or backward difference schemes were used at boundaries. Temporal gradients were calculated using a backward difference scheme, where the gradient value is equal to the performance loss value at the current moment minus the performance loss value at the previous moment, divided by the time step, which was set to 1 hour. A method for calculating the gradient amplitude was established, where the gradient amplitude is equal to the square root of the sum of the squares of the spatial gradient components. The gradient direction is calculated using the inverse tangent function, with an angle range of 0° to 360° and an accuracy of 0.1°. A gradient field visualization technique was used to generate a gradient distribution cloud map, which uses color coding to represent gradient intensity, with blue indicating low gradient areas and red indicating high gradient areas. The number of color levels was set to 256. A gradient anomaly detection algorithm was established, using a statistical distribution-based anomaly detection method. The gradient anomaly threshold was set at the gradient mean plus twice the standard deviation. Regions exceeding the threshold were marked as high-risk degradation areas. A Gaussian filter algorithm was used for gradient data smoothing, with the filter kernel standard deviation set to 0.02 and the filter window size to a 5×5 grid cell. A smoothed gradient field was obtained through convolution. A gradient convergence analysis was established, evaluating convergence characteristics using the rate of change of the gradient modulus. Convergence was determined when the rate of change of the gradient modulus was less than 1% / h over 10 consecutive time steps. Four levels of insulation layer performance loss gradient were established: 0% / h to 5% / h for low loss gradients, 5% / h to 15% / h for medium loss gradients, 15% / h to 30% / h for high loss gradients, and exceeding 30% / h for extremely high loss gradients. Gradient data was stored in a sparse matrix format, storing only non-zero gradient values ​​and their corresponding coordinate indices. This achieved a compression ratio of 85%. The storage format was a coordinate-compressed row format, consisting of three basic components: a value array, a column index array, and a row pointer array. The gradient analysis result output includes information such as maximum gradient value, gradient distribution map, risk area identification, time evolution trend, etc.

[0044] Step S234 includes the following steps: Based on the temperature rise inflection point growth rate difference and the temperature rise growth convergence range, the thermal cycle stress fluctuation simulation analysis of the basic characteristics of the line insulation material is carried out to obtain the material thermal stress fluctuation data; Based on the material thermal stress fluctuation data, the thermal shear plasticity offset of the basic characteristics of the line insulation material is simulated to obtain the thermal shear plasticity offset data; The material plasticizer loss index is quantified based on the material thermal stress fluctuation data and the thermal shear plasticity offset data to obtain the material plasticizer loss index; Based on the thermal shear plasticity offset data and the material plasticizer loss index, the basic characteristics of the line insulation material are numerically analyzed for the thermal degradation evolution to obtain the numerical evolution law of the material thermal degradation.

[0045] In an embodiment of the present invention, based on the temperature rise inflection point growth rate difference and the temperature rise growth convergence interval data, a thermal cycle stress fluctuation simulation algorithm is established, and the finite element thermal stress analysis method is used to process the stress response of the insulating material under periodic temperature load. The simulation grid is set to a hexahedral unit structure with a unit size of 0.02mm×0.02mm×0.02mm, a total number of grids of 125,000, and a number of nodes of 140,608. The temperature load is applied in the form of a sine wave function, with a temperature amplitude range of 25°C to 180°C, a period set to 4h corresponding to the actual operating conditions, and an initial value of the phase angle of 0°. The material constitutive relationship adopts a linear elastic model with an elastic modulus of 2.5GPa, a Poisson's ratio of 0.34, and a linear thermal expansion coefficient. , density 1420kg / m³. The thermal stress calculation is solved by displacement finite element method. The stiffness matrix is ​​obtained by Gaussian integration. The number of integration points is 8. The solver uses preconditioned conjugate gradient method. The convergence tolerance is set to Stress fluctuation analysis uses the rainflow counting method to statistically analyze stress cycle characteristics, setting the stress cycle threshold to 1 MPa. Statistical parameters include the number of cycles, stress amplitude, and average stress. A stress fluctuation spectrum analysis algorithm is established, using fast Fourier transform to convert the time-domain stress signal into a frequency-domain representation. The sampling frequency is 100 Hz, the analysis frequency range is 0.01 Hz to 50 Hz, and the frequency resolution is 0.01 Hz. The stress fluctuation data storage format includes fields such as timestamp, stress component, stress amplitude, and fluctuation frequency, with a data sampling interval of 36 seconds.

[0046] Based on the thermal stress fluctuation data of the material, a thermal shear plasticity offset simulation algorithm is established, and the elastic-plastic finite element method is used to analyze the plastic deformation behavior of the insulating material under shear stress. The material yield criterion is set to the von Mises yield criterion, the yield stress is 75 MPa, the tangent modulus is 500 MPa, the hardening law adopts the isotropic hardening model, and the hardening parameters are determined by uniaxial tensile tests. The shear stress is applied in a pure shear loading along the main direction of the material, and the shear strain rate is set to , loading time 3600s corresponds to 1h continuous loading. The plastic deformation calculation adopts the return mapping algorithm, and the iterative convergence criterion is that the L2 norm of the plastic strain increment is less than The maximum number of iterations was 200. A method for calculating the plastic offset was established. The offset is equal to the equivalent strain value of the plastic strain tensor. The equivalent plastic strain calculation formula is the square root of half the sum of the squares of the differences between the three principal plastic strains. An incremental method was used to track the evolution of plastic deformation. The time increment was set to 10 seconds, and the nonlinear equations were solved using Newton-Raphson iteration within each increment. Statistical analysis of shear plastic offset data includes parameters such as maximum offset, offset rate, and offset direction. The offset unit is dimensionless strain value, with an accuracy of 5 decimal places. A method for calculating offset cumulative damage was established. The damage variable is defined as the ratio of the current plastic strain to the failure plastic strain. The failure plastic strain was determined to be 0.025 through material tensile testing. The direction cosine method was used to analyze the offset directional characteristics. The offset direction was determined by the principal direction of the plastic strain, with an angle accuracy of 0.1°. The thermal shear plastic offset data was stored in a tensor format, containing information such as the strain tensor components, principal strain values, and principal directions. Symmetric tensor storage optimization was used for data compression, reducing storage space by 50%.

[0047] Based on the material thermal stress fluctuation data and thermal shear plasticity offset data, a quantitative algorithm for plasticizer loss index is established, and the diffusion dynamics theory is used to analyze the migration law of plasticizer molecules in the polymer matrix. The plasticizer diffusion coefficient is set to , activation energy 120kJ / mol, and the temperature dependence of the diffusion coefficient is described by the Arrhenius relation. A finite difference solution format for the diffusion equation is established, the central difference format is used for spatial discretization, the forward Euler format is used for time discretization, the grid size is 0.001mm, the time step is 0.1s, and the stability condition requires the diffusion number to be less than 0.5. The plasticizer loss is calculated by boundary flux integration. The boundary conditions are set as first-class boundary conditions, and a boundary concentration of 0 indicates a complete loss state. The loss index is defined as the ratio of the current plasticizer concentration to the initial concentration. The initial concentration is set to 15% mass fraction. The loss index ranges from 0 to 1, and the accuracy is maintained at 4 decimal places. A multi-factor coupling analysis framework is established, considering the synergistic effects of temperature, stress, and deformation on plasticizer loss. The coupling coefficient is determined by orthogonal experimental design, with a temperature influence coefficient of 1.8, a stress influence coefficient of 0, and a deformation influence coefficient of 1.2. The Monte Carlo method was used to account for the randomness of the erosion process. Random variables included parameters such as the diffusion coefficient, boundary conditions, and initial concentration. The parameter distribution type was a lognormal distribution with a coefficient of variation of 15%, and 5000 simulations were performed. The temporal evolution of the erosion index was analyzed using an exponential decay function fit. The fitting parameters included the initial value, decay constant, and equilibrium value, with an R² goodness-of-fit requirement of greater than 0.98. A spatial distribution analysis of the erosion index was established, and contour maps of the erosion index were generated using the Kriging interpolation method. The interpolation radius was 0.5 mm, and a Gaussian weighting function was used.

[0048] Based on the thermal shear plasticity offset data and the material plasticizer loss index, a numerical evolution analysis algorithm for material thermal degradation is established, and a multi-scale modeling method is used to correlate the molecular scale degradation mechanism with the macroscopic performance change. The degradation evolution equation is set as a first-order ordinary differential equation system, and the state variables include molecular chain breakage rate, crystallinity, dielectric constant, etc. The evolution equation is solved using the Runge-Kutta fourth-order method with a time step of 0.01h. The evolution time span of 8760h corresponds to 1 year. The evolution of the molecular chain breakage rate adopts a chain reaction kinetic model, and the reaction rate constant , reaction order 2, activation energy 150kJ / mol. The crystallinity change is described by the Avrami equation, the Avrami index is 2.8, and the crystallization rate constant is , equilibrium crystallinity 35%. The dielectric constant evolution adopts linear degradation model, and the degradation rate , with an initial value of 3.5 and a final stable value of 2.8. A degradation coupling coefficient matrix was established. The matrix dimensions were 3×3, corresponding to the interactions between the three degradation parameters. Matrix elements were determined through regression of experimental data, with the main diagonal elements set to 1.0 and the off-diagonal elements ranging from 0.1 to 0.3. A sensitivity analysis was used to assess the influence of various factors on the degradation rate. Sensitivity coefficients were calculated using the finite difference method, with the perturbation amplitude set to 1% of the nominal parameter value. The degradation evolution pattern was represented using a piecewise function and divided into three phases: an initial rapid degradation phase, a mid-term stable degradation phase, and a late accelerated degradation phase. The durations of each phase were 720 h, 6480 h, and 1560 h, respectively. A confidence interval analysis of the degradation prediction was established, using bootstrap resampling to estimate prediction uncertainty. The number of resamplings was 1000, the confidence level was 95%, and the confidence interval width was kept within 10% of the predicted value.

[0049] Step S24 includes the following steps: Step S241: performing a high-frequency current oscillation waveform analysis based on the overload multiple evolution data to obtain a high-frequency current oscillation waveform; performing a steep and disordered oscillation analysis on the high-frequency current oscillation waveform to obtain a steep and disordered oscillation waveform; Step S242: performing multi-frequency transient mutation rate coupling on the steep disordered oscillation waveform to obtain multi-frequency transient mutation coupling data; Step S243: Deducing the corresponding failure boundary of the insulation layer based on the insulation layer performance loss gradient according to the multi-frequency transient mutation coupling data to obtain insulation layer failure boundary data; Step S244: performing a winding short circuit multi-factor superposition risk probability calculation based on the multi-frequency transient mutation coupling data and the insulation layer failure boundary data, thereby obtaining the winding short circuit risk probability; Step S245: Calculate the motor failure probability based on the winding short circuit risk probability to obtain the motor failure calculation probability.

[0050] In this embodiment of the present invention, a bandpass filter is used to extract high-frequency current oscillation components based on overload multiple evolution data. The filter design uses an 8th-order elliptic filter with a passband frequency range of 800Hz to 8000Hz, a passband ripple of 0.1dB, a stopband attenuation of 60dB, and a transition band width of 100Hz. The overload multiple evolution data is preprocessed with a sampling frequency of 20kHz and a data length of 65,536 sampling points corresponding to a 3.2768s time window. A Hamming window is used as the window function to reduce spectral leakage. The high-frequency oscillation waveform is extracted using a Hilbert transform to obtain an analytical signal. The instantaneous amplitude is obtained from the analytical signal modulus, and the instantaneous frequency is obtained from the phase derivative. Phase unwrapping uses the unwrap algorithm to handle 2π jumps. An algorithm for detecting steep disordered oscillations was developed using a wavelet transform multiresolution analysis method. The Daubechies-10 wavelet was selected as the wavelet basis function, and the decomposition layer was 8, with the frequency bands of each layer ranging from 625Hz to 1250Hz, 1250Hz to 2500Hz, 2500Hz to 5000Hz, and 5000Hz to 10000Hz, respectively. Steepness was determined using wavelet coefficient amplitude gradient analysis, with the gradient threshold set at 5 times the wavelet coefficient mean. High-gradient events with a duration of less than 0.01s were labeled as steep oscillations. Disorder was determined using an approximate entropy algorithm to quantify signal complexity. The sequence length was set to 256 data points, the pattern dimension was 2, and the tolerance parameter was 0.15 times the signal standard deviation. Approximate entropy values ​​greater than 1.2 were labeled as disordered oscillations. Compressed sensing technology is used to store data for steep, disordered oscillation waveforms. The discrete cosine transform (DCT) is used as the sparse transform basis, and a Gaussian random matrix is ​​used as the measurement matrix. The compression ratio is 75%, and the reconstruction algorithm uses the orthogonal matching pursuit algorithm. The reconstruction error is kept within 1% of the original signal energy. The extraction of oscillation waveform feature parameters includes statistical features such as peak factor, skewness, kurtosis, and energy concentration. The peak factor is defined as the ratio of the signal peak value to the effective value. The skewness reflects the asymmetry of the signal probability distribution, and the kurtosis reflects the sharpness of the signal probability distribution. The energy concentration is quantified using Shannon entropy to determine the concentration of energy in the frequency domain.

[0051] Multi-frequency transient mutation detection is performed on steep, disordered oscillatory waveforms. A change-point detection algorithm is used to identify the moment of sudden change in the signal's statistical characteristics. The detection method is based on the principle of cumulative sum control charts. The decision function is a cumulative sum sequence, the reference value is set to the signal mean, and the decision interval is three times the signal standard deviation. The transient mutation rate is defined as the ratio of the mutation amplitude to the mutation duration. The mutation amplitude is obtained by the difference in signal amplitude before and after the mutation, and the mutation duration is determined by the length of time that the mutation continuously exceeds the detection threshold. The rate unit is V / s. A multi-frequency analysis framework is established, and the oscillatory waveform is divided into six sub-bands based on frequency. The band division is equally logarithmically spaced, with the center frequencies of each band being 1000Hz, 1500Hz, 2250Hz, 3375Hz, 5063Hz, and 7594Hz, respectively, and the bandwidth is 1 / 3 octave of the center frequency. An instantaneous frequency tracking algorithm was used to analyze transient characteristics within each frequency band. The tracking algorithm is based on a Kalman filter. The state vector contains the instantaneous frequency and the frequency rate of change. The process noise covariance is 0.01 Hz², and the observation noise covariance is 0.1 Hz². For transient mutation rate coupling analysis, a cross-correlation function was used to calculate the time-delay correlation between different frequency bands. The cross-correlation function has a length of 1024 sampling points. The normalized correlation coefficient ranges from -1 to 1. An absolute value of the correlation coefficient greater than 0.7 indicates strong coupling, 0.3 to 0.7 indicates moderate coupling, and less than 0.3 indicates weak coupling. A coupling strength matrix was constructed with a dimension of 6×6, corresponding to the pairwise coupling relationships between the six frequency bands. Matrix elements represent the maximum cross-correlation coefficients between the corresponding frequency bands, and a diagonal element of 1 indicates autocorrelation. Principal component analysis was used to extract the main features of the coupling data. The principal component directions were obtained through eigenvalue decomposition of the covariance matrix. The number of principal components with a cumulative contribution rate of 95% was considered the effective dimension, and the principal component scores were used as coupling eigenvectors. Multi-frequency transient mutation coupling data was stored in a sparse matrix format.

[0052] Based on multi-frequency transient mutation coupling data, an insulation failure boundary deduction algorithm was established, and a failure boundary discriminant function was constructed using a support vector machine classification method. The training dataset consisted of 5,000 historical insulation failure cases. The feature vector dimension was 18 dimensions, containing six parameters each for coupling strength, average mutation rate, and maximum mutation rate in six frequency bands. The labels were binary variables representing failure and normal states. The support vector machine kernel function used a radial basis function, with the kernel parameter γ set to 0.1 and the penalty parameter C set to 100. Hyperparameters were optimized using a grid search method with 10-fold cross-validation, achieving a classification accuracy of 94.5%. The failure boundary deduction algorithm used a contour tracing algorithm to search for a hypersurface where the classification decision function was equal to zero in the 18-dimensional feature space. The contour step size was 0.01, and the tracing accuracy was 0.001. A failure probability density function was estimated, using a kernel density estimation method to fit the probability distribution of failure samples in the feature space. A Gaussian kernel was used as the kernel function, and the bandwidth parameter was automatically selected using the Silverman criterion, with a value of 0.25. The insulation failure boundary data is represented using an implicit surface equation. The equation coefficients are obtained through least squares fitting. The fitting order is third-order and includes 84 coefficients, including linear, quadratic, and cubic terms. The sum of squared residuals is less than 0.01. The dynamic evolution of the failure boundary is tracked using the level set method. The initial value of the level set function is the signed distance function, and the evolution equation is the Hamilton-Jacobi equation. The numerical solution uses the high-resolution WENO scheme with a CFL number of 0.5 and a time step of 0.001 s. A quantitative analysis of the failure boundary uncertainty is established. The Monte Carlo method is used to propagate the uncertainty of the input parameters. The number of random samplings is 10,000. The input parameter distribution type is uniform, and the distribution interval is ±20% of the nominal value. The failure boundary data is validated using an independent test set consisting of 1,000 data sets. The prediction accuracy is 92.8%, the false positive rate is 3.2%, and the false negative rate is 4.0%.

[0053] According to the multi-frequency transient mutation coupling data and insulation layer failure boundary data, a probability calculation algorithm for the multi-factor superposition risk of winding short circuit is established, and the Bayesian network method is used to model the causal relationship and conditional dependency between multiple factors. The Bayesian network nodes include five parent nodes such as insulation aging, overload current, temperature rise abnormality, mechanical vibration, and ambient humidity, and one child node for winding short circuit. The node state is discretized, and the three states of each node correspond to low, medium, and high risk levels. The conditional probability table is determined jointly by expert knowledge and historical fault data. The conditional probability of insulation aging for short circuit is low risk 0.02, medium risk 0.15, and high risk 0.45. The conditional probability of overload current for short circuit is low risk 0.01, medium risk 0.08, and high risk 0.35. The variational inference algorithm is used for probability inference. The variational distribution adopts the mean field approximation. The variational parameters are optimized by the coordinate ascent variational inference algorithm. The convergence criterion is that the variational lower bound change is less than The maximum number of iterations was 500. Multi-factor cumulative effect modeling used a Copula function to describe the correlation structure between factors. The Copula function type was ClaytonCopula, and the correlation parameters were obtained through maximum likelihood estimation. A parameter value of 2.3 indicated a moderate positive correlation. Risk probability calculations were performed using Monte Carlo simulations, with 50,000 simulations. Each simulation sampled from the marginal distribution of each factor. Correlated samples were generated using the Copula function, and the proportion of samples falling into the failure region was calculated as the risk probability estimate. A sensitivity analysis framework was established, and the Sobol index was used to quantify the contribution of each factor to the total risk. The first-order Sobol index reflects the main effect of a single factor, and the total Sobol index reflects the total effect including interactions. The index was calculated using the Saltelli sampling method, with a sample size of 100,000. The confidence interval for the winding short-circuit risk probability was estimated using the bootstrap method, with 1,000 resampling cycles, a 95% confidence level, and a confidence interval width of 0.03. The Markov Chain Monte Carlo method is used to analyze the time evolution of risk probability. The state transition matrix is ​​obtained from historical data statistics. The transition probability is updated every hour, and the prediction time window is 168 hours, corresponding to one week. Winding short-circuit risk probability data is stored in a time series database with a time resolution of 1 hour. Data fields include probability values, confidence intervals, key risk factors, and sensitivity indexes.

[0054] Based on the winding short-circuit risk probability data, an algorithm for calculating the motor failure probability is established, and the fault tree analysis method is used to model the failure logic relationship of the motor system. The top event of the fault tree is motor failure, the middle events include three subsystem failures such as electrical failure, mechanical failure, and thermal failure, and the bottom events include 15 basic failure modes such as winding short-circuit, bearing failure, insulation aging, rotor bar breakage, and stator core failure. The logic gate types include OR gate, AND gate, and k / n voting gate. Winding short-circuit and insulation aging are connected through AND gates to indicate that they will cause electrical failure only when they occur simultaneously. Bearing failure and rotor bar breakage are connected through OR gates to indicate that either one of them will cause mechanical failure. The failure probability calculation adopts the minimum cut set method, and the fault tree logic expression is simplified through Boolean algebra. There are 27 minimum cut sets, each of which contains 1 to 4 basic events. There are 8 minimum cut sets for single-point failure, 12 minimum cut sets for double-point failure, 6 minimum cut sets for three-point failure, and 1 minimum cut set for four-point failure. The basic event failure rate data comes from the reliability database. The winding short-circuit failure rate , bearing failure rate , insulation aging failure rate , rotor bar failure rate , stator core failure rate The formula for calculating the system failure probability is 1 minus the product of all minimum cut set reliabilities. The reliability of each minimum cut set is equal to 1 minus the product of the failure probabilities of the basic events within the minimum cut set. The failure probability of a basic event is equal to 1 minus the negative failure rate of the natural constant multiplied by the operating time. Importance analysis is used to assess the impact of each basic event on system failure. The Fussell-Vesely importance reflects the contribution of the basic event to the system unreliability, the Birnbaum importance reflects the sensitivity of changes in the failure probability of the basic event to the system failure probability, and the critical importance comprehensively considers the current state and impact of the basic event. The time-dependent analysis of the motor failure probability is modeled using the Weibull distribution. The shape parameter β is determined by regression of failure data to 2.1, representing wear failure characteristics. The scale parameter η is set to 50,000 hours to represent the characteristic life. The location parameter γ is set to 0, indicating that the calculation starts at time zero. The Fisher information matrix method is used to calculate the variance-covariance matrix of the parameter estimates for the failure probability prediction confidence interval, with a confidence level of 90%. The prediction time range is 8760 hours, corresponding to a one-year operating cycle. The quality control of the motor failure probability data adopts the cross-validation method. The historical failure data are divided into a training set and a validation set, with the training set ratio being 70% and the validation set ratio being 30%. The prediction accuracy evaluation indicator is the root mean square error.

[0055] Step S3 includes the following steps: Step S31: normalizing the calculated probability of motor failure to obtain a normalized probability of motor failure; Step S32: assigning an overload current risk level to the current overload spectrum diagram based on the normalized probability of motor failure to obtain an overload current risk level; Step S33: constructing an equipment fault diagnosis framework based on the normalized probability of motor failure and the overload current risk level; Step S34: Send the device fault diagnosis framework to the control terminal of the electromechanical device to perform diagnosis of the electromechanical device fault.

[0056] As an example of the present invention, refer to Figure 3 As shown, in this example, step S3 includes: Step S31: normalizing the calculated probability of motor failure to obtain a normalized probability of motor failure; In the embodiment of the present invention, the motor failure calculation probability is normalized, and the minimum-maximum normalization algorithm is used to map the probability value to the range of 0 to 1. The normalization formula is: the normalized probability is equal to the current failure probability minus the minimum failure probability divided by the maximum failure probability minus the minimum failure probability. The numerical range of the failure probability is determined by statistical analysis of historical fault data, and the minimum failure probability is set to For normal equipment operation, the maximum failure probability is set to 0.95, corresponding to an impending failure state. A probabilistic data preprocessing process is established. First, outlier detection is performed on the raw failure probability data. The 3σ criterion is used to identify abnormal data points that exceed the mean plus or minus three standard deviations. The proportion of outliers is controlled within 2%, and data exceeding this proportion is recalculated. A moving average filter algorithm is used to smooth the failure probability time series. The filter window length is set to 12 data points, corresponding to a 12-hour time span. The filter weights adopt a Gaussian distribution with a standard deviation of 0.3, a center weight of 0.4, and an edge weight decay of 0.05. A normalized parameter adaptive update mechanism is established. The minimum and maximum parameters are updated every 24 hours using an exponentially weighted moving average with a smoothing coefficient of 0.1. The new parameter value is equal to 0.1 times the current statistical value plus 0.9 times the historical parameter value. The Kolmogorov-Smirnov test was used to evaluate the quality of normalized data and verify the uniformity of data distribution. A test statistic D value less than 0.05 indicated a good normalization effect, and a p-value greater than 0.05 indicated that the data distribution conformed to the uniform distribution hypothesis. A normalized data storage format was established using the IEEE 754 double-precision floating-point format, with numerical accuracy maintained at 15 significant digits, 8 bytes of storage space per data point, and compression using the gzip algorithm with a compression ratio of 65%. The normalized probability data was verified using an inverse normalization test, which reversely mapped the normalized results back to the original numerical range and calculated the inverse normalization error, requiring the relative error to be less than 0.1% and the absolute error to be less than .

[0057] Step S32: assigning an overload current risk level to the current overload spectrum diagram based on the normalized probability of motor failure to obtain an overload current risk level; In one embodiment of the present invention, an algorithm for assigning overload current risk levels is established based on normalized motor failure probability data. A hierarchical clustering method is used to classify frequency points in the current overload spectrum according to risk level. Risk level classification criteria are set: normalized probabilities of 0 to 0.25 correspond to a low risk level, 0.25 to 0.5 correspond to a medium-low risk level, 0.5 to 0.75 correspond to a medium-high risk level, and 0.75 to 1.0 correspond to a high risk level. A frequency-probability mapping relationship is established, and a linear interpolation algorithm is used to map the discrete failure probability data to a continuous frequency domain. Interpolation nodes are frequency sampling points in the spectrum, and cubic spline interpolation is used. The boundary condition is a natural boundary condition, i.e., the second-order derivative is zero. The spectrum data is clustered by risk level using the k-means clustering algorithm. The number of clusters is set to 4, corresponding to the four risk levels. Initial cluster centers are selected using the k-means++ algorithm. The distance metric is the Euclidean distance. The convergence criterion is that the change in cluster center is less than 0.001, and the maximum number of iterations is 100. A risk level color coding system was established: low risk levels were coded using green RGB values ​​(0, 255, 0), medium-low risk levels were coded using yellow RGB values ​​(255, 255, 0), medium-high risk levels were coded using orange RGB values ​​(255, 165, 0), and high risk levels were coded using red RGB values ​​(255, 0, 0). The risk level assignment process used a nearest neighbor classification method. For each pixel in the spectrum graph, the distance between it and each cluster center was calculated, and the pixel was assigned to the risk level corresponding to the cluster center closest to it. A risk level smoothing algorithm was established, using a median filter with a 3×3 pixel window size to eliminate noise points in the risk level distribution. After filtering, a connected domain analysis was performed, merging isolated regions with an area smaller than 10 pixels into adjacent regions of the primary risk level. Overload current risk level data was stored using run-length encoding, representing consecutive pixels of the same risk level by their starting position and length. The storage compression ratio was 80%, and the decompression time was less than 10 ms. The statistical analysis of risk level distribution includes geometric features such as the area ratio of each level, boundary length, and shape factor. The area ratio is calculated as the number of pixels in each level divided by the total number of pixels multiplied by 100%. The boundary length is obtained through the edge detection algorithm, and the shape factor is defined as the square of the perimeter divided by the area.

[0058] Step S33: constructing an equipment fault diagnosis framework based on the normalized probability of motor failure and the overload current risk level; In this embodiment of the present invention, an equipment fault diagnosis architecture is established based on the normalized probability of motor failure and overload current risk level data. This architecture utilizes a layered system design, encompassing five functional layers: data acquisition, signal processing, feature extraction, fault identification, and decision output. The data acquisition layer is responsible for acquiring sensor signals such as current, voltage, temperature, and vibration in real time. It operates at a sampling frequency of 20kHz, a data bit depth of 16 bits, and uses 24 sensors. Data transmission utilizes the CAN bus protocol, a baud rate of 1Mbps, a standard frame format, and an 8-byte data frame length. The signal processing layer implements digital filtering, spectrum analysis, and feature transformation. The filter utilizes an IIR digital filter with an 8th-order filter and a cutoff frequency of 10kHz. Spectral analysis utilizes the FFT algorithm with a transform length of 4096 points, a Hanning window function, and a 50% overlap ratio. The feature extraction layer extracts multidimensional feature vectors from the processed signal, including time-domain features, frequency-domain features, and time-frequency features. Time-domain features include six parameters: mean, variance, skewness, kurtosis, crest factor, and crest factor. Frequency-domain features include four parameters: power spectral density, spectral center of gravity, spectral width, and spectral skewness. Time-frequency features are obtained via short-time Fourier transform (SFT) with a window length of 512 points and an overlap rate of 75%. The fault identification layer uses a support vector machine classifier for fault pattern recognition. The classifier training data consists of 8,000 sets of normal and faulty samples, with a 32-dimensional feature dimension, a radial basis function kernel with a kernel parameter of 0.1, and a penalty parameter of 50. The classification accuracy is 96.5%. The decision output layer generates a diagnostic report and maintenance recommendations based on the fault identification results. The report format is an XML structured document and contains fields such as fault type, severity, occurrence time, impact range, and recommended actions. Maintenance recommendations are generated using an expert system rule base consisting of 150 if-then rules. The architecture's communication interface utilizes a RESTful API design, supporting HTTP / HTTPS protocols and JSON data format. The interface response time is less than 100ms, with 1,000 concurrent connections and OAuth 2.0 authentication. The diagnostic architecture is deployed using containerization technology, with an Alpine Linux-based container image. The image size is 150MB, the memory usage is 256MB, and the CPU utilization is less than 20%. It supports horizontal scaling and load balancing.

[0059] Step S34: Send the device fault diagnosis framework to the control terminal of the electromechanical device to perform diagnosis of the electromechanical device fault.

[0060] In an embodiment of the present invention, the device fault diagnosis architecture is sent to the electromechanical equipment control terminal via industrial Ethernet, and the FTP file transfer protocol is used for remote deployment of the architecture program. The FTP server address is 192.168.1.100, the port number is 21, the user name is admin, and the password is stored using MD5 hash encryption. The architecture program is packaged in a tar.gz compressed package, which contains components such as executable files, configuration files, dependency libraries, and documents. The compressed package size is 85MB, and the transmission time is about 30s. After the transmission is completed, an MD5 check is performed to ensure file integrity. The control terminal hardware platform uses an ARM Cortex-A9 processor with a main frequency of 1.2GHz, 1GB DDR3 memory, 32GB eMMC storage, an operating system Linux kernel version 4.19, and a real-time extension PREEMPT_RT patch. The architecture program is installed using an automated script with the scripting language bash. The installation steps include dependency checking, permission setting, service registration, configuration initialization, etc. The installation time is about 5 minutes, and the installation log is recorded in the / var / log / install.log file. The diagnostic service is started using the systemd service manager. The service name is motor-diagnosis, the service type is Type=forking, the startup command is ExecStart= / usr / bin / motor-diagnosis, and the restart policy is Restart=always. The service status is queried using the systemctl status command. A remote monitoring mechanism is established to regularly report the diagnostic service's operating status via the SNMP protocol. SNMP version v3, port 161, community string public, and a custom MIB library containing object identifiers such as service status, CPU usage, memory usage, and diagnostic counts. The fault diagnosis execution cycle is set to automatically trigger the complete diagnostic process every 10 minutes. Emergency diagnosis is triggered by an external interrupt signal, with an interrupt response time of less than 1ms. Diagnostic results are stored in a local SQLite database with a file size limit of 100MB. A diagnostic result upload mechanism is established to send diagnostic results to the cloud-based monitoring platform via the MQTT protocol.

[0061] The present invention further provides a data analysis-based electromechanical equipment fault diagnosis system for executing the data analysis-based electromechanical equipment fault diagnosis method described above. The data analysis-based electromechanical equipment fault diagnosis system includes: The overload status analysis module is used to extract the historical operation log of the motor from the control terminal of the electromechanical equipment; analyze the current overload status of the historical operation log and construct a current overload spectrum diagram; A failure probability calculation module is used to analyze the insulation layer performance loss gradient based on the current overload spectrum to obtain the insulation layer performance loss gradient; and to calculate the motor failure probability based on the insulation layer performance loss gradient to obtain the motor failure calculation probability; A fault diagnosis architecture building module is used to assign an overload current risk level to the current overload spectrum diagram based on the motor failure calculation probability to obtain the overload current risk level; build an equipment fault diagnosis architecture according to the overload current risk level; and send the equipment fault diagnosis architecture to the control terminal of the electromechanical equipment to perform the diagnosis of the electromechanical equipment fault.

[0062] The foregoing description is intended only to provide specific embodiments of the present invention, which will enable those skilled in the art to understand and implement the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not intended to be limited to the embodiments shown herein, but is to be construed in the widest possible manner consistent with the principles and novel features disclosed herein.

Claims

1. A method for diagnosing electromechanical equipment faults based on data analysis, characterized in that: The following steps are involved: Step S1: extracting the historical operation log of the motor from the control terminal of the electromechanical equipment; analyzing the current overload status of the historical operation log and constructing a current overload spectrum diagram; Step S2: performing insulation layer performance loss gradient analysis based on the current overload spectrum to obtain the insulation layer performance loss gradient; performing motor failure probability calculation based on the insulation layer performance loss gradient to obtain the motor failure calculation probability; Step S3: assigning an overload current risk level to the current overload spectrum diagram based on the motor failure calculation probability to obtain the overload current risk level; Build equipment fault diagnosis architecture based on overload current risk level; The device fault diagnosis framework is sent to the control terminal of the electromechanical device to perform diagnosis of the electromechanical device fault.

2. The electromechanical equipment fault diagnosis method based on data analysis according to claim 1, characterized in that: Step S1 includes the following steps: Step S11: extracting the historical operation log of the motor from the control terminal of the electromechanical equipment; Step S12: extracting the motor overload operation state from the motor's historical operation log to obtain the overload operation state; Step S13: performing current overload state analysis on the overload operation state and constructing a current overload spectrum diagram.

3. The electromechanical equipment fault diagnosis method based on data analysis according to claim 1, characterized in that: Step S2 includes the following steps: Step S21: performing overload multiple evolution analysis based on the current overload spectrum diagram to obtain overload multiple evolution data; Step S22: Calculating the line temperature rise cumulative gain index based on the overload multiple evolution data to obtain the line temperature rise cumulative index; Step S23: performing insulation layer performance loss gradient analysis based on the line temperature rise cumulative index to obtain the insulation layer performance loss gradient; Step S24: Calculating the motor failure probability based on the overload multiple evolution data and the insulation layer performance loss gradient to obtain the motor failure calculation probability.

4. The electromechanical equipment fault diagnosis method based on data analysis according to claim 3, characterized in that: Step S22 includes the following steps: Step S221: performing an incremental multiple change rate analysis on the overload multiple evolution data to obtain an incremental trend multiple change rate; Step S222: performing mutation time series integration on the overload multiple evolution data according to the increasing trend multiple change rate, thereby obtaining the increasing mutation integral multiple; Step S223: performing aggregation of equivalent calorific value gains per unit time according to the incremental mutation integration multiple to obtain a calorific value gain aggregation sequence; Step S224: performing recursive superposition regression analysis on the heat gain series per unit time to generate a line temperature rise accumulation curve; performing exponential fitting processing on the line temperature rise accumulation curve to obtain temperature rise accumulation exponential fitting data; Step S225: Calculating the line temperature rise cumulative gain index based on the temperature rise cumulative index fitting data to obtain the line temperature rise cumulative index.

5. The electromechanical equipment fault diagnosis method based on data analysis according to claim 3, characterized in that: Step S23 includes the following steps: Step S231: Obtain basic properties of line insulation materials; Step S232: constructing a continuous temperature rise variation curve based on the line temperature rise cumulative index, and then calculating the average difference of the increasing rate of the temperature rise inflection point to obtain the increasing rate difference of the temperature rise inflection point; Step S233: performing slope growth convergence analysis on the continuous temperature rise change curve according to the temperature rise inflection point growth rate difference to obtain the temperature rise growth convergence range; Step S234: performing a numerical evolution analysis of thermal degradation of the basic characteristics of the line insulation material based on the temperature rise inflection point growth rate difference and the temperature rise growth convergence interval to obtain a numerical evolution law of thermal degradation of the material; Step S235: performing insulation layer performance loss gradient analysis according to the numerical evolution law of material thermal degradation to obtain the insulation layer performance loss gradient.

6. The electromechanical equipment fault diagnosis method based on data analysis according to claim 5, characterized in that: Step S234 includes the following steps: Based on the temperature rise inflection point growth rate difference and the temperature rise growth convergence interval, the thermal cycle stress fluctuation simulation analysis of the basic characteristics of the line insulation material is carried out to obtain the material thermal stress fluctuation data; Based on the material thermal stress fluctuation data, the thermal shear plasticity offset of the basic characteristics of the line insulation material is simulated to obtain the thermal shear plasticity offset data; The material plasticizer loss index is quantified based on the material thermal stress fluctuation data and the thermal shear plasticity offset data to obtain the material plasticizer loss index; Based on the thermal shear plasticity offset data and the material plasticizer loss index, the basic characteristics of the line insulation material are numerically analyzed for the thermal degradation evolution to obtain the numerical evolution law of the material thermal degradation.

7. The electromechanical equipment fault diagnosis method based on data analysis according to claim 3, characterized in that: Step S24 includes the following steps: Step S241: performing a high-frequency current oscillation waveform analysis based on the overload multiple evolution data to obtain a high-frequency current oscillation waveform; performing a steep and disordered oscillation analysis on the high-frequency current oscillation waveform to obtain a steep and disordered oscillation waveform; Step S242: performing multi-frequency transient mutation rate coupling on the steep disordered oscillation waveform to obtain multi-frequency transient mutation coupling data; Step S243: Deducing the corresponding failure boundary of the insulation layer based on the insulation layer performance loss gradient according to the multi-frequency transient mutation coupling data to obtain insulation layer failure boundary data; Step S244: performing a winding short circuit multi-factor superposition risk probability calculation based on the multi-frequency transient mutation coupling data and the insulation layer failure boundary data, thereby obtaining the winding short circuit risk probability; Step S245: Calculate the motor failure probability based on the winding short circuit risk probability to obtain the motor failure calculation probability.

8. The electromechanical equipment fault diagnosis method based on data analysis according to claim 1, characterized in that: Step S3 includes the following steps: Step S31: normalizing the calculated probability of motor failure to obtain a normalized probability of motor failure; Step S32: assigning an overload current risk level to the current overload spectrum diagram based on the normalized probability of motor failure to obtain an overload current risk level; Step S33: constructing an equipment fault diagnosis framework based on the normalized probability of motor failure and the overload current risk level; Step S34: Send the device fault diagnosis framework to the control terminal of the electromechanical device to perform diagnosis of the electromechanical device fault.

9. A mechanical and electrical equipment fault diagnosis system based on data analysis, characterized in that: For executing the electromechanical equipment fault diagnosis method based on data analysis as claimed in claim 1, the electromechanical equipment fault diagnosis system based on data analysis comprises: The overload status analysis module is used to extract the historical operation log of the motor from the control terminal of the electromechanical equipment; analyze the current overload status of the historical operation log and construct a current overload spectrum diagram; A failure probability calculation module is used to analyze the insulation layer performance loss gradient based on the current overload spectrum to obtain the insulation layer performance loss gradient; and to calculate the motor failure probability based on the insulation layer performance loss gradient to obtain the motor failure calculation probability; A fault diagnosis architecture building module is used to assign an overload current risk level to the current overload spectrum diagram based on the motor failure calculation probability to obtain the overload current risk level; build an equipment fault diagnosis architecture according to the overload current risk level; and send the equipment fault diagnosis architecture to the control terminal of the electromechanical equipment to perform the diagnosis of the electromechanical equipment fault.

Images