Intelligent failure trend pre-judgment method for inductive device

Abstract

The invention relates to the technical field of inductance devices, and discloses an intelligent failure trend pre-judgment method for an inductance device, which comprises the following steps: S1, data acquisition: acquiring multi-dimensional electrical parameters representing the state of the inductance device in real time in the production and working process of the inductance device, S2, data preprocessing, and S3, pre-judging the failure trend of the inductance device. Carrying out filtering and normalization processing on the collected data; s3, establishing a trend model; s4, contrastive analysis and grade judgment are carried out, inductor state parameter data which are collected and preprocessed in real time are input into the trend prediction model, and S5, dynamic adjustment and updating are carried out. According to the method, a monitoring mode which only depends on a single temperature sensor is abandoned, the multi-dimensional electrical parameters representing the state of the inductor device are collected in real time, the temperature and humidity of the environment are collected at the same time, the situation that the failure of the inductor device is possibly caused by multiple factors is fully considered, and the failure pre-judgment accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of inductor devices, and specifically to an intelligent failure trend prediction method for inductor devices. Background Technique

[0002] The intelligent working principle of inductor devices is the deep integration of traditional electromagnetic induction effect with modern sensing technology, data analysis, and adaptive control systems. In the prior art, it mostly relies on manual inspection, regular detection, or single-parameter threshold judgment, and realizes protection by attaching a temperature sensor, but it can only be triggered after reaching the high-temperature threshold, resulting in some significant delays.

[0003] After retrieval, the patent with the Chinese patent number CN108365746B discloses a high-gain bidirectional four-phase DC-DC converter based on coupled inductors and its control method, which includes a four-phase coupled inductor. The inductances of each phase of the coupled inductor are L1 to L4, and the coupling method is positive and negative coupling, that is, positive coupling between odd phases and even phases of the coupled inductor, and negative coupling between odd and even phases. There are eight power switches S1 to S8, and five capacitors CL, C1, C2, CH1, and CH2. The present invention has a higher voltage conversion ratio, and its voltage conversion ratio can be increased to four times that of the traditional bidirectional four-phase DC-DC converter.

[0004] After retrieval, the patent with the Chinese patent number CN108492958B discloses a series-type multi-phase interleaved coupled inductor structure and its control method, which is composed of several series-connected multi-phase coupled inductor units. Each multi-phase coupled inductor unit winds N winding coils on a single magnetic core, and the N winding coils are sequentially positively and negatively coupled, where N is the number of phases of the coupled inductor; the magnetic core of each multi-phase coupled inductor unit adopts an annular magnetic ring structure; the winding directions of the odd-phase windings on each magnetic core are the same, the winding directions of the even-phase windings are the same, and the winding directions of the odd-phase and even-phase windings are opposite, so as to realize the sequential positive and negative coupling of each winding on each magnetic core.

[0005] The above-mentioned patent relies solely on a single temperature sensor to monitor the operating state of the inductor device. For example, when the temperature reaches a high-temperature threshold, a protection mechanism is triggered. However, this method ignores that the failure of the inductor device may be caused by a combination of multiple factors, such as changes in parameters like current, voltage, and magnetic field intensity. It is difficult to comprehensively and accurately predict the failure trend of the inductor device based on a single parameter alone, and false positives or false negatives are likely to occur. Additionally, the method with a fixed threshold lacks flexibility and adaptability and cannot be dynamically adjusted according to the actual operating conditions and historical data of the inductor device. For example, under different operating loads and environmental conditions, the normal operating parameter range of the inductor device may change, but the fixed threshold cannot adapt to this change, which may lead to false triggering of protection during normal operation or failure to detect potential problems in a timely manner at the initial stage of failure due to the parameters not reaching the threshold. Based on this, the present invention designs an intelligent failure trend prediction method for inductor devices to solve the above problems. Summary of the Invention

[0006] The purpose of the present invention is to provide an intelligent failure trend prediction method for inductor devices, which solves the problems of false positives, false negatives, and inability to provide timely protection in the background technology.

[0007] To solve the above technical problems, the present invention provides the following technical solutions:

[0008] An intelligent failure trend prediction method for inductor devices includes the following steps:

[0009] Step S1, data collection. During the production and operation of the inductor device, multi-dimensional electrical parameters characterizing the state of the inductor device are collected in real time, including the working current value, the voltage value at both ends, the self-inductance coefficient, the mutual inductance coefficient, and the winding resistance value. At the same time, the temperature and humidity parameters of the environment where the inductor device is located are collected.

[0010] Step S2, data preprocessing. The collected data is filtered and normalized to remove noise interference and outliers, making the data format unified and the dimensions consistent.

[0011] Step S3, establish a trend model. Based on historical production and test data, a trend prediction model of the state parameters of the inductor device is constructed using machine learning algorithms.

[0012] Step S4, comparative analysis and level determination. The preprocessed data of the state parameters of the inductor device collected in real time is input into the trend prediction model, and the deviation degree between the current parameters and the model prediction values is calculated. When the deviation degree exceeds the preset dynamic threshold range, an early warning mechanism is triggered. According to the magnitude and change rate of the deviation degree, combined with the importance and application scenario of the inductor device, different early warning levels are divided to provide a basis for subsequent maintenance decisions.

[0013] Step S5, dynamic adjustment and update. Periodically retrain and optimize the trend prediction model using newly collected data, enabling the model to adapt to the slow drift of the inductor device performance and changes in the working environment. Meanwhile, adjust the warning threshold and the warning level classification criteria according to the actual operation situation.

[0014] Preferably, in step S1, the following steps are included:

[0015] Step S1.1, acquisition of the working current value. Using a high-precision current sensor, within the working frequency range of the inductor device, the instantaneous value of the working current is collected in real time, and the waveform characteristic parameters of the current are recorded simultaneously. Among them, the form factor is calculated by the formula Kf = Irms / Iavg, where Irms is the root mean square value of the current and Iavg is the average value of the current.

[0016] Step S1.2, acquisition of the voltage value at both ends. Using a high-precision voltage sensor, synchronized with the acquisition frequency of the working current, the instantaneous voltage value at both ends of the inductor device is obtained in real time, and the characteristic parameters such as the peak value, root mean square value, and form factor of the voltage waveform are also extracted.

[0017] Step S1.3, measurement of the self-inductance coefficient. During the production process of the inductor device, using a dedicated inductance test instrument and the AC bridge method, according to the formula L = (V × t) / (I × N), the self-inductance coefficient of each inductor device is accurately measured, where V is the applied voltage, t is the charging time, I is the charging current, and N is the number of turns of the coil. During the working process, the self-inductance coefficient is estimated and monitored regularly through an on-line measurement circuit using the frequency response analysis method.

[0018] Preferably, in step S1, the following steps are further included:

[0019] Step S1.4, measurement of the mutual inductance coefficient. For polyphase coupled inductor devices, during the production test stage, using a dual-trace oscilloscope and a signal generator, combined with the mutual inductance voltage formula of the coupled inductor M = (E2 × t) / (I1 × N2), where E2 is the induced voltage of the secondary coil, t is the change time, I1 is the current of the primary coil, and N2 is the number of turns of the secondary coil. In the working state, by monitoring the changes in the mutual inductance voltage and current between each phase, the mutual inductance coefficient is estimated in real time using numerical analysis methods.

[0020] Step S1.5, Acquisition of winding resistance value. The four-wire measurement method is adopted, and a high-precision resistance measuring instrument is used to monitor the change of the resistance value of the inductor winding in real time. During the working process, a weak detection current is applied to both ends of the winding, the voltage drop on it is measured, and the winding resistance is calculated according to Ohm's law R = V / I. At the same time, considering the influence of temperature on the winding resistance, the measurement result is temperature-compensated according to the formula R = R0[1 + α(T - T0)], where R0 is the resistance value at the reference temperature T0, α is the temperature coefficient of resistance, and T is the current ambient temperature.

[0021] Preferably, in the said step S2, the following steps are included:

[0022] Step S2.1, Filtering process. The wavelet transform filtering algorithm is adopted to perform multi-scale decomposition on the collected electrical parameter data to remove high-frequency noise and power frequency interference. Specifically, the signal is decomposed into 3 layers of wavelets, a suitable wavelet basis function is selected, threshold processing is performed on the high-frequency part, and then wavelet reconstruction is carried out to obtain the filtered signal;

[0023] Step S2.1, Normalization process. The normalization method is adopted to map the data to the interval [0, 1]. The normalization formula is X_normalized = (X - X_min) / (X_max - X_min), where X is the original data, and X_min and X_max are the minimum and maximum values in the data sample respectively.

[0024] Preferably, in the said step S3, the following steps are included:

[0025] Step S3.1, Data preprocessing and feature extraction. Before establishing the trend model, the historical data is preprocessed, including data cleaning, missing value filling, and outlier processing; the interpolation method based on time series is used to fill the missing values, and the box plot method is used to detect and process the outliers; feature extraction is performed on the preprocessed data, and time domain features, frequency domain features, and time-frequency domain features are extracted to form a feature vector as the input of the model;

[0026] Step S3.2, Model training and verification. The extracted feature vectors and the corresponding target values are combined to form a training data set, and the long short-term memory network algorithm is used for model training; after the training is completed, the test set is used to evaluate the model, and the indicators of the mean square error, mean absolute error, and determination coefficient of the model are calculated to evaluate the prediction performance of the model.

[0027] Preferably, in the said step S4, the following steps are included:

[0028] Step S4.1, Deviation Degree Calculation and Early Warning Triggering: Input the preprocessed state parameter data of the inductor device collected in real time into the trend prediction model to calculate the deviation degree between the current parameters and the model prediction values. The weighted Mahalanobis distance method is used to consider the correlation and weight differences between parameters, and the calculation formula is D = sqrt[(Z - Z_pred)^TΣ^{-1}W(Z - Z_pred)], where Z is the parameter vector collected and preprocessed in real time, Z_pred is the parameter vector predicted by the trend prediction model, Σ is the parameter covariance matrix, and W is the weight matrix. The weights are determined according to the sensitivity of each parameter to failure. When the deviation degree D exceeds the preset dynamic threshold range, the early warning mechanism is triggered.

[0029] Preferably, in step S4, the following steps are included:

[0030] Step S4.2, Early Warning Level Judgment and Decision Support: According to the magnitude and change rate of the deviation degree, combined with the importance and application scenario of the inductor device, different early warning levels are divided. The specific judgment criteria are as follows:

[0031] Criterion S1: D < D_low, and the change rate |ΔD| < ΔD_low, indicating that the state of the inductor device is normal and it can operate normally;

[0032] Criterion S2: D_low ≤ D < D_mid, or ΔD_low ≤ |ΔD| < ΔD_mid, indicating that the inductor device has a slight abnormality and it is recommended to conduct further monitoring;

[0033] Criterion S3: D_mid ≤ D < D_high, or ΔD_mid ≤ |ΔD| < ΔD_high, indicating that the inductor device has a relatively obvious abnormality and it is recommended to conduct a detailed inspection in the near future;

[0034] Criterion S4: D ≥ D_high, or |ΔD| ≥ ΔD_high, indicating that the inductor device has a serious failure risk and it is necessary to immediately stop the machine for maintenance or replacement.

[0035] Preferably, in step S5, the following steps are included:

[0036] Step S5.1, Model Update and Optimization Strategy: Regularly retrain and optimize the trend prediction model using the newly collected data. The update period is dynamically determined according to the mean time between failures of the inductor device and the actual operating conditions. In the initial stage of equipment operation or during the stage of frequent operating condition changes, and in the stable stage of equipment operation; the transfer learning method is adopted, and the parameters of the original model are used as the initialization parameters of the new model, and only the newly added data is trained and used to improve the training efficiency;

[0037] Step S5.2, Dynamically adjust the warning threshold and the grading criteria. According to the actual operation conditions and historical warning records, dynamically adjust the warning threshold and the warning grading criteria; adopt the Bayesian statistical method, combine the newly collected data with prior knowledge, update the probability distribution of the deviation degree, and thus adjust the dynamic threshold range.

[0038] Preferably, it further includes a data storage and management step:

[0039] Step S6, Data storage and management. Store the collected raw data, preprocessed data, trend prediction results, deviation degree calculation results, and warning information into the database. The database adopts a distributed storage architecture; at the same time, establish a data indexing and query mechanism to facilitate the quick retrieval and analysis of historical data, classify and store the data, and file it according to the model number, batch number, and production date information of the inductor components.

[0040] Preferably, this method displays the real-time state parameters, trend prediction curves, deviation degree, and warning level information of the inductor components in a graphical manner. The interface includes a real-time data display area, a trend chart area, a warning information prompt area, and an operation control area. Intuitively present the change trends and prediction results of each parameter through visual charts, which is used for operators to quickly understand the operation status of the equipment.

[0041] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0042] 1. The present invention abandons the monitoring method that only relies on a single temperature sensor, and instead collects multi-dimensional electrical parameters that characterize the state of the inductor components in real time, including the working current value, the voltage value at both ends, the self-inductance coefficient, the mutual inductance coefficient, and the winding resistance value. At the same time, collect the temperature and humidity of the environment, fully considering the situation that the failure of the inductor components may be caused by multiple factors jointly, greatly reducing the risk of misjudgment or missed judgment due to only monitoring a single parameter, and improving the accuracy of failure prediction.

[0043] 2. The present invention uses the newly collected data to retrain and optimize the trend prediction model regularly, enabling it to adapt to the slow drift of the inductor component performance and the changes in the working environment, and dynamically adjust the warning threshold and the warning grading criteria according to the actual operation conditions. In this way, it overcomes the limitation of the traditional fixed threshold that cannot adapt to different working loads and environmental conditions, avoiding false triggering of protection under normal working conditions and the situation that potential problems cannot be detected in time due to the parameters not reaching the threshold in the initial stage of failure.

[0044] 3. In the present invention, by using a machine learning algorithm to construct a trend prediction model for the state parameters of an inductor device, and adopting the weighted Mahalanobis distance method to comprehensively consider the correlation and weight differences between parameters to calculate the deviation degree, it is possible to more scientifically and accurately evaluate the deviation degree between the current state and the predicted state of the inductor device; at the same time, different warning levels are divided according to the magnitude and change rate of the deviation degree, as well as the importance and application scenarios of the inductor device, providing a reference basis for subsequent maintenance decisions. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] Figure 1 is the overall flowchart of the intelligent failure trend prediction method for the inductor device of the present invention;

[0046] Figure 2 is the detailed flowchart of data acquisition of the present invention;

[0047] Figure 3 is the flowchart for establishing the trend prediction model of the present invention;

[0048] Figure 4 is the flowchart for determining the warning level of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0049] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0050] Embodiment 1;

[0051] Please refer to Figures 1 - 4 , in the embodiment of the present invention, an intelligent failure trend prediction method for an inductor device includes the following steps:

[0052] Step S1, data acquisition. During the production and operation of the inductor device, multi-dimensional electrical parameters representing the state of the inductor device are collected in real time, including the working current value, the voltage value at both ends, the self-inductance coefficient, the mutual inductance coefficient, and the winding resistance value. At the same time, the temperature and humidity parameters of the environment where the inductor device is located are collected;

[0053] Step S2, data preprocessing. The collected data is filtered and normalized to remove noise interference and outliers, so that the data format is unified and the dimension is consistent;

[0054] Step S3, establishing a trend model. Based on historical production and test data, a trend prediction model for the state parameters of the inductor device is constructed by using a machine learning algorithm;

[0055] Step S4, Comparative Analysis and Grade Judgment: The preprocessed inductor device status parameter data collected in real time is input into the trend prediction model to calculate the deviation degree between the current parameters and the model prediction values. When the deviation degree exceeds the preset dynamic threshold range, the early warning mechanism is triggered; according to the magnitude and change rate of the deviation degree, combined with the importance and application scenarios of the inductor device, different early warning levels are divided to provide a basis for subsequent maintenance decisions;

[0056] Step S5, Dynamic Adjustment and Update: Regularly retrain and optimize the trend prediction model using newly collected data so that the model can adapt to the slow drift of the inductor device performance and the changes in the working environment. At the same time, adjust the early warning threshold and the early warning level division criteria according to the actual operation situation.

[0057] In step S1, the following steps are included:

[0058] Step S1.1, Acquisition of Working Current Value: Using a high-precision current sensor, within the working frequency range of the inductor device, the instantaneous value of the working current is collected in real time, and at the same time, the waveform characteristic parameters of the current are recorded. Among them, the form factor is calculated by the formula Kf = Irms / Iavg, where Irms is the effective current value and Iavg is the average current value;

[0059] Step S1.2, Acquisition of Voltage Values at Both Ends: Using a high-precision voltage sensor, synchronized with the acquisition frequency of the working current, the instantaneous voltage value at both ends of the inductor device is obtained in real time, and the characteristic parameters such as the peak value, root mean square value, and form factor of the voltage waveform are also extracted;

[0060] Step S1.3, Measurement of Self-Inductance Coefficient: During the production process of the inductor device, using a special inductance testing instrument and the alternating current bridge method, according to the formula L = (V × t) / (I × N), the self-inductance coefficient of each inductor device is accurately measured, where V is the applied voltage, t is the charging time, I is the charging current, and N is the number of turns of the coil; during the working process, the self-inductance coefficient is estimated and monitored regularly through an on-line measurement circuit using the frequency response analysis method;

[0061] Step S1.4, Measurement of Mutual Inductance Coefficient: For multi-phase coupled inductor devices, during the production test stage, using a dual-trace oscilloscope and a signal generator, combined with the mutual inductance voltage formula of the coupled inductor M = (E2 × t) / (I1 × N2), where E2 is the induced voltage of the secondary coil, t is the change time, I1 is the current of the primary coil, and N2 is the number of turns of the secondary coil; in the working state, by monitoring the changes in the mutual inductance voltage and current between each phase, the mutual inductance coefficient is estimated in real time using numerical analysis methods;

[0062] Step S1.5, acquisition of winding resistance value. The four-wire measurement method is adopted, and a high-precision resistance measuring instrument is used to monitor the change of the resistance value of the inductor winding in real time. During the working process, a weak detection current is applied across the winding, the voltage drop across it is measured, and the winding resistance is calculated according to Ohm's law R = V / I. At the same time, considering the influence of temperature on the winding resistance, the measurement result is temperature-compensated according to the formula R = R0[1 + α(T - T0)], where R0 is the resistance value at the reference temperature T0, α is the resistance temperature coefficient, and T is the current ambient temperature.

[0063] In step S2, the following steps are included:

[0064] Step S2.1, filtering process. The wavelet transform filtering algorithm is adopted to perform multi-scale decomposition on the collected electrical parameter data to remove high-frequency noise and power frequency interference. Specifically, the signal is decomposed into 3 layers of wavelets, a suitable wavelet basis function is selected, threshold processing is performed on the high-frequency part, and then wavelet reconstruction is carried out to obtain the filtered signal.

[0065] Step S2.1, normalization process. The normalization method is adopted to map the data to the interval [0, 1]. The normalization formula is X_normalized = (X - X_min) / (X_max - X_min), where X is the original data, and X_min and X_max are the minimum and maximum values in the data sample respectively.

[0066] In step S3, the following steps are included:

[0067] Step S3.1, data preprocessing and feature extraction. Before establishing the trend model, the historical data is preprocessed, including data cleaning, missing value filling, and outlier processing. The interpolation method based on time series is used to fill the missing values, and the box plot method is used to detect and process the outliers. Feature extraction is performed on the preprocessed data, and time domain features, frequency domain features, and time-frequency domain features are extracted to form a feature vector as the input of the model.

[0068] Step S3.2, model training and verification. The extracted feature vectors and the corresponding target values are combined to form a training data set, and the long short-term memory network algorithm is used for model training. After training, the test set is used to evaluate the model, and the indicators of mean square error, mean absolute error, and determination coefficient of the model are calculated to evaluate the prediction performance of the model.

[0069] In step S4, the following steps are included:

[0070] Step S4.1, Deviation Calculation and Warning Triggering: Input the preprocessed inductor device status parameter data collected in real time into the trend prediction model to calculate the deviation between the current parameter and the model prediction value. The weighted Mahalanobis distance method is used to consider the correlation and weight differences between parameters, and the calculation formula is D = sqrt[(Z - Z_pred)^TΣ^{-1}W(Z - Z_pred)], where Z is the parameter vector collected and preprocessed in real time, Z_pred is the parameter vector predicted by the trend prediction model, Σ is the parameter covariance matrix, and W is the weight matrix. The weights are determined according to the sensitivity of each parameter to failure. When the deviation D exceeds the preset dynamic threshold range, the warning mechanism is triggered.

[0071] Step S4.2, Warning Level Judgment and Decision Support: According to the magnitude and change rate of the deviation, combined with the importance and application scenario of the inductor device, different warning levels are divided. The specific judgment criteria are as follows: Criterion S1: D < D_low and the change rate |ΔD| < ΔD_low, indicating that the inductor device is in normal condition and can operate normally; Criterion S2: D_low ≤ D < D_mid or ΔD_low ≤ |ΔD| < ΔD_mid, indicating that the inductor device has a slight abnormality and it is recommended to conduct further monitoring; Criterion S3: D_mid ≤ D < D_high or ΔD_mid ≤ |ΔD| < ΔD_high, indicating that the inductor device has an obvious abnormality and it is recommended to conduct a detailed inspection in the near future; Criterion S4: D ≥ D_high or |ΔD| ≥ ΔD_high, indicating that the inductor device has a serious failure risk and it is necessary to immediately stop the machine for maintenance or replacement.

[0072] The working principle of the embodiment of the present invention is as follows: The system realizes the all-round perception of the working state of the inductor device by integrating a high-precision sensor array and an embedded monitoring module. Among them, the current and voltage parameters adopt synchronous acquisition technology, combined with the dynamic calculation of the form factor (Kf), to accurately reflect the instantaneous load characteristics of the device; the online estimation of the self-inductance coefficient and mutual inductance coefficient is realized through frequency response analysis and coupled voltage modeling, breaking through the limitations of traditional offline detection; at the same time, the environmental temperature and humidity parameters are combined with the temperature compensation algorithm of the winding resistance value to eliminate the influence of external interference on the core electrical parameters, forming a multi-dimensional data representation system.

[0073] The system uses the wavelet transform hierarchical filtering algorithm to process the original data, and reconstructs the signal through three-layer wavelet decomposition to effectively separate high-frequency noise and power frequency interference. The normalization process adopts the dynamic interval mapping method, and adjusts the X_min and X_max thresholds according to the real-time data distribution to avoid the distortion influence of historical extreme values on the current data normalization. In the feature extraction stage, the time-frequency domain joint analysis technology is introduced to extract time-frequency features including the current harmonic distortion rate and the voltage transient response slope, and construct a feature vector with physical interpretability.

[0074] The prediction model based on LSTM adopts a dual-channel input structure. The first channel inputs the time-series electrical parameter feature vectors, and the second channel inputs the environmental parameters and device aging coefficients. The model dynamically adjusts the weights of historical memory and current input through a gating unit to capture the non-linear degradation law of inductive devices. During the training process, a transfer learning strategy is adopted to pre-train the model base using historical production data, and then the model is fine-tuned under specific working conditions through online incremental learning.

[0075] The early warning judgment module adopts an improved weighted Mahalanobis distance algorithm to quantify the coupling relationship between parameters through the covariance matrix Σ, and introduces a sensitivity weight matrix W (such as the winding resistance weight coefficient is 0.35 and the self-inductance coefficient is 0.28), making the calculation of the deviation degree D more in line with the actual failure mechanism. The dynamic threshold setting adopts a sliding window statistical method to automatically adjust the boundary values of D_low, D_mid, and D_high according to the prediction error distribution in the recent 30 days. The early warning level division integrates the analysis of the derivative of the change rate. When |ΔD / Δt| exceeds the set threshold, a cross-level early warning is triggered to effectively identify the risk of sudden failure.

[0076] The system is built-in with a model update engine. By comparing the residual distribution of the prediction deviation and the measured data, three update modes are triggered: parameter fine-tuning (deviation < 5%), structure optimization (5% ≤ deviation < 15%), and full model reconstruction (deviation ≥ 15%). The environment adaptive module continuously monitors the drift error of the temperature and humidity sensors. When it detects a decrease in the sensor accuracy, it automatically switches to the backup sensing node and performs data fusion through Kalman filtering.

[0077] Embodiment 2;

[0078] Please refer to Figures 1 - 4 , in the embodiment of the present invention, in step S5, it includes the following steps:

[0079] Step S5.1, model update and optimization strategy: Regularly retrain and optimize the trend prediction model using newly collected data. The update period is dynamically determined according to the mean time between failures of the inductive device and the actual operating conditions. In the initial stage of equipment operation or during the stage of frequent working condition changes, and during the stable stage of equipment operation; adopt the transfer learning method, use the parameters of the original model as the initialization parameters of the new model, and only train the new data and use it to improve the training efficiency;

[0080] Step S5.2, dynamic adjustment of early warning thresholds and level division criteria: Dynamically adjust the early warning thresholds and early warning level division criteria according to the actual operating conditions and historical early warning records; adopt the Bayesian statistical method, combine the newly collected data and prior knowledge, update the probability distribution of the deviation degree, and thus adjust the dynamic threshold range.

[0081] It also includes steps of data storage and management:

[0082] Step S6, data storage and management: Store the collected raw data, preprocessed data, trend prediction results, deviation degree calculation results, and warning information in a database, and the database adopts a distributed storage architecture; at the same time, establish a data indexing and query mechanism to facilitate the rapid retrieval and analysis of historical data, classify and store the data, and file it according to the model number, batch number, and production date information of the inductor device.

[0083] This method displays the real-time status parameters, trend prediction curves, deviation degrees, and warning level information of the inductor device in a graphical manner. The interface includes a real-time data display area, a trend chart area, a warning information prompt area, and an operation control area. The change trends and prediction results of each parameter are intuitively presented through visual charts, which are used for operators to quickly understand the operating status of the equipment.

[0084] The working principle of the embodiment of the present invention is as follows: The system adopts a hierarchical update strategy, dynamically adjusts the model update period according to the equipment operation stage. In the initial stage of equipment operation (the first 3000 hours) or the stage with frequent working condition fluctuations, full-volume data retraining is performed every 72 hours; in the stable operation stage (MTBF≥10^5 hours), the incremental learning mode is adopted to update the model parameters weekly. The transfer learning module retains the weights of the feature extraction layer of the historical model through the parameter freezing technology, and only fine-tunes the fully connected layer to ensure the knowledge inheritance efficiency under the difference in the distribution of new and old data, and the convergence speed of transfer training is increased by 42%. The model optimization engine is built with a residual analysis module, and when the prediction deviation exceeds 5% continuously for 5 times, it triggers a structure reconstruction, and automatically generates a hybrid network architecture of LSTM and GRU to adapt to the non-linear degradation characteristics.

[0085] The warning threshold generation module establishes a three-layer probability distribution model. The basic layer fits the device failure time characteristics based on the Weibull distribution of historical data. The middle layer uses a Markov chain to describe the temporal correlation of parameter deviations. The top layer updates the prior distribution by fusing real-time monitoring data through Bayesian inference. The data archiving engine establishes a three-level index according to the SN code (the structure is YYMMDD-XXXXX). The first level establishes a B+ tree index according to the device model, the second level establishes a hash partition according to the production batch, and the third level constructs a time wheel storage structure based on the production date, and the query response time is shortened to within 50ms. The data cleaning module integrates an improved DBSCAN algorithm, identifies outliers through an adaptive ε parameter, and the cleaning efficiency is 3.8 times higher than that of traditional methods.

[0086] The graphical interface uses WebGL 3D rendering technology to construct a predictive trajectory sphere model containing a four-dimensional state space, and real-time displays the projection positions of parameter points in the safe area, warning area, and danger area. The trend chart area integrates the wavelet denoising algorithm and the ARIMA prediction curve for superimposed display, and supports a data refresh rate of 0.1 s level. The early warning prompt module adopts a multi-channel warning strategy. The primary warning triggers the yellow breathing light effect, the intermediate warning starts the synchronous sound and light alarm, and the advanced warning directly pushes the maintenance work order to the MES system.

[0087] Embodiment 3;

[0088] Please refer to Figures 1 - 4 , in the embodiment of the present invention, a specific embodiment is provided. The Hall current sensor captures the working current waveform in real time, synchronously records the root mean square value (Irms) and the average value (Iavg), and calculates the form factor Kf = Irms / Iavg; the voltage sensor synchronously collects the voltage peak value and the root mean square value, and extracts the transient response slope feature through coupling impedance analysis; the acoustic emission sensor array is arranged at the joint of the inductor winding and the magnetic core to detect the stress wave signal, and extracts the instantaneous amplitude variance (0.02 - 0.15 V2) and the main frequency band energy ratio (≥85%) of 8 IMF components through HHT transform; the environmental monitoring unit integrates a PT100 temperature sensor (±0.5 °C) and a capacitive humidity sensor (±3% RH), updates the environmental parameters every 10 seconds, and corrects the winding resistance value by combining the temperature compensation algorithm.

[0089] Perform 4-layer db4 wavelet decomposition on the electrical signal, use the improved SURE threshold algorithm to process the high-frequency noise, and the signal-to-noise ratio after reconstruction is increased to 42 dB; based on the extreme value mapping of the sliding window (window length 30 min, step size 5 min), eliminate the dimension difference caused by the working condition fluctuation; extract 27-dimensional time-frequency features, including the current harmonic distortion rate (abnormal mark is triggered when THD≥5%), the winding temperature rise rate (threshold 0.5 °C / min), and the acoustic emission energy entropy value.

[0090] Input 23-dimensional parameters (including the root mean square value of current, voltage ripple coefficient, etc.), 300 nodes in the LSTM hidden layer, and the initial bias of the forgetting gate is 0.7; input 14-dimensional parameters (including the kurtosis of the main frequency band, the energy mutation rate of IMF3, etc.), 200 nodes in the GRU hidden layer, and the weight of the update gate is 0.68; dynamically allocate the feature weights, and combine the transfer learning strategy: use 10,000 groups of accelerated aging data (temperature cycle -40 °C to 150 °C) in the pre-training stage, and adopt the AdamW optimizer in the online learning stage.

[0091] Deviation calculation is based on the improved weighted Mahalanobis distance \(D = \sqrt{(Z - Z_{pred})^T\Sigma^{-1}W(Z - Z_{pred})}\). The covariance matrix \(\Sigma\) is updated every 5 minutes through sequential Monte Carlo sampling. In the weight matrix \(W\), the weight of the winding resistance is 0.35 and the self-inductance coefficient is 0.28. The Bayesian-particle filter fusion algorithm is adopted. The benchmark threshold \(D_{threshold}=\mu\pm k\sigma\), where the value of \(k\) is dynamically configured according to the device level (for industrial grade \(k = 2.58\), for automotive grade \(k = 3.29\)). The sliding window statistically analyzes the error distribution in the recent 30 days. When the change acceleration \(\alpha=\Delta D / \Delta t^2\geq0.5\), a three-level early warning is directly triggered, and the emergency protection mechanism is synchronously started.

[0092] Through the above scheme, this embodiment realizes an early warning 240 hours in advance before the inductor fails (accuracy rate: 92.3%), with a false alarm rate of ≤2.1%, and the timeliness is improved by 3.8 times compared with the traditional method.

[0093] Working principle: During the production and operation of the inductor device, multi-dimensional electrical parameters characterizing the state of the inductor device are collected in real time, including the working current value, the voltage value at both ends, the self-inductance coefficient, the mutual inductance coefficient, and the winding resistance value. At the same time, the temperature and humidity parameters of the environment where the inductor device is located are collected. Then, the collected data is filtered and normalized to remove noise interference and outliers, so that the data format is unified and the dimension is consistent.

[0094] Based on historical production and test data, a trend prediction model of the inductor device state parameters is constructed using machine learning algorithms. The data of the inductor device state parameters collected and preprocessed in real time is input into the trend prediction model, and the deviation degree between the current parameters and the model prediction value is calculated. When the deviation degree exceeds the preset dynamic threshold range, the early warning mechanism is triggered. According to the magnitude and change rate of the deviation degree, combined with the importance and application scenario of the inductor device, different early warning levels are divided to provide a basis for subsequent maintenance decisions.

[0095] In addition, the trend prediction model is retrained and optimized regularly using newly collected data, so that the model can adapt to the slow drift of the inductor device performance and the changes in the working environment. At the same time, the early warning threshold and the early warning level division standard are adjusted according to the actual operation situation. Through the real-time monitoring and analysis of the multi-dimensional parameters of the inductor device and the trend prediction model constructed by combining machine learning algorithms, the present invention realizes the intelligent prediction of the inductor device failure trend and improves the accuracy and timeliness of the prediction.

[0096] Although the embodiments of the present invention have been shown and described, for those of ordinary skill in the art, it can be understood that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirits of the present invention. The scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent failure trend prediction method for an inductor device, characterized in that Including the following steps: Step S1, data acquisition: During the production and operation of the inductor device, multi-dimensional electrical parameters characterizing the state of the inductor device are collected in real time, including the working current value, the voltage value at both ends, the self-inductance coefficient, the mutual inductance coefficient, and the winding resistance value. At the same time, the temperature and humidity parameters of the environment where the inductor device is located are collected; Step S2, data preprocessing: The collected data is filtered and normalized to remove noise interference and outliers, so that the data format is unified and the dimension is consistent; Step S3, establishing a trend model: Based on historical production and test data, a trend prediction model of the inductor device state parameters is constructed using machine learning algorithms; Step S4, comparative analysis and grade determination: The preprocessed inductor device state parameter data collected in real time is input into the trend prediction model, and the deviation degree between the current parameter and the model prediction value is calculated. When the deviation degree exceeds the preset dynamic threshold range, an early warning mechanism is triggered; According to the magnitude and change rate of the deviation degree, combined with the importance and application scenario of the inductor device, different early warning levels are divided to provide a basis for subsequent maintenance decisions; Step S5, dynamic adjustment and update: Regularly use the newly collected data to retrain and optimize the trend prediction model, so that the model can adapt to the slow drift of the inductor device performance and the changes in the working environment. At the same time, adjust the early warning threshold and the early warning level division standard according to the actual operation situation.

2. The intelligent failure trend prediction method for an inductor device according to claim 1, characterized in that In the said step S1, it includes the following steps: Step S1.1, acquisition of the working current value: Using a high-precision current sensor, within the working frequency range of the inductor device, the instantaneous value of the working current is collected in real time, and the waveform characteristic parameters of the current are recorded at the same time. Among them, the form factor is calculated by the formula Kf = Irms / Iavg, where Irms is the effective current value and Iavg is the average current value; Step S1.2, acquisition of the voltage value at both ends: Using a high-precision voltage sensor, synchronized with the acquisition frequency of the working current, the instantaneous voltage value at both ends of the inductor device is obtained in real time, and the characteristic parameters of the peak value, root mean square value and form factor of the voltage waveform are also extracted; Step S1.3, measurement of the self-inductance coefficient: During the production process of the inductor device, using a special inductance test instrument and adopting the AC bridge method, according to the formula L = (V × t) / (I × N), the self-inductance coefficient of each inductor device is accurately measured, where V is the applied voltage, t is the charging time, I is the charging current, and N is the number of turns of the coil; During the operation process, the self-inductance coefficient is estimated and monitored regularly through an on-line measurement circuit using the frequency response analysis method.

3. The intelligent failure trend prediction method for an inductor device according to claim 1, wherein In the said step S1, it also includes the following steps: Step S1.4, measurement of the mutual inductance coefficient: For multi-phase coupled inductor devices, during the production test stage, using a dual-trace oscilloscope and a signal generator, combined with the mutual inductance voltage formula of the coupled inductor M = (E2 × t) / (I1 × N2), where E2 is the induced voltage of the secondary coil, t is the change time, I1 is the current of the primary coil, and N2 is the number of turns of the secondary coil; In the working state, by monitoring the mutual inductance voltage and current changes between each phase, the mutual inductance coefficient is estimated in real time using numerical analysis methods; Step S1.5, Acquisition of winding resistance value. The four-wire measurement method is adopted, and a high-precision resistance measuring instrument is used to monitor the change of the resistance value of the inductor winding in real time. During the working process, a weak detection current is applied across the winding, the voltage drop across it is measured, and the winding resistance is calculated according to Ohm's law R = V / I. At the same time, considering the influence of temperature on the winding resistance, the measurement result is temperature-compensated according to the formula R = R0[1 + α(T - T0)], where R0 is the resistance value at the reference temperature T0, α is the temperature coefficient of resistance, and T is the current ambient temperature.

4. The intelligent failure trend prediction method for an inductor device according to claim 1, characterized in that In the said step S2, the following steps are included: Step S2.1, Filtering process. The wavelet transform filtering algorithm is adopted to perform multi-scale decomposition on the collected electrical parameter data to remove high-frequency noise and power frequency interference. Specifically, the signal is decomposed into 3 layers of wavelets, a suitable wavelet basis function is selected, threshold processing is performed on the high-frequency part, and then wavelet reconstruction is carried out to obtain the filtered signal. Step S2.1, Normalization process. The normalization method is adopted to map the data to the interval [0, 1]. The normalization formula is X_normalized = (X - X_min) / (X_max - X_min), where X is the original data, and X_min and X_max are the minimum and maximum values in the data sample respectively.

5. The intelligent failure trend prediction method of an inductor component according to claim 1, characterized in that In the said step S3, the following steps are included: Step S3.1, Data preprocessing and feature extraction. Before establishing the trend model, the historical data is preprocessed, including data cleaning, missing value filling, and outlier processing. The interpolation method based on time series is used to fill the missing values, and the box plot method is used to detect and process the outliers. Feature extraction is performed on the preprocessed data, and time domain features, frequency domain features, and time-frequency domain features are extracted to form a feature vector as the input of the model. Step S3.2, Model training and verification. The extracted feature vectors and the corresponding target values are combined to form a training data set, and the long short-term memory network algorithm is used for model training. After the training is completed, the test set is used to evaluate the model, and the indicators of the mean square error, mean absolute error, and coefficient of determination of the model are calculated to evaluate the prediction performance of the model.

6. The intelligent failure trend prediction method for an inductor device according to claim 1, wherein In the said step S4, the following steps are included: Step S4.1, Deviation degree calculation and warning trigger. The state parameter data of the inductor device collected and preprocessed in real time is input into the trend prediction model, and the deviation degree between the current parameter and the model prediction value is calculated. The weighted Mahalanobis distance method is used to consider the correlation and weight differences between the parameters. The calculation formula is D = sqrt[(Z - Z_pred)^TΣ^{-1}W(Z - Z_pred)], where Z is the parameter vector collected and preprocessed in real time, Z_pred is the parameter vector predicted by the trend prediction model, Σ is the parameter covariance matrix, and W is the weight matrix. The weight is determined according to the sensitivity of each parameter to failure. When the deviation degree D exceeds the preset dynamic threshold range, the warning mechanism is triggered.

7. The intelligent failure trend prediction method for an inductor device according to claim 1, wherein In the said step S4, the following steps are included: Step S4.2, warning level determination and decision support, divides different warning levels according to the size and change rate of the deviation, combined with the importance and application scenarios of the inductor components. The specific determination criteria are as follows: Standard S1: D <D_low,且变化速率|ΔD|<ΔD_low,表示电感器件状态正常,可正常运行; Standard S2: D_low≤D <D_mid,或ΔD_low≤|ΔD|<ΔD_mid,表示电感器件出现轻微异常,建议进行进一步监测; Standard S3: D_mid≤D <D_high,或ΔD_mid≤|ΔD|<ΔD_high,表示电感器件出现较明显异常,建议近期进行详细检测; Standard S4: D ≥ D_high, or |ΔD| ≥ ΔD_high, indicates that the inductor is at serious risk of failure and requires immediate shutdown for inspection or replacement.

8. The intelligent failure trend prediction method for an inductor device according to claim 1, characterized in that, The step S5 includes the following steps: Step S5.1: Model update and optimization strategy. Regularly retrain and optimize the trend prediction model using newly collected data. The update cycle is dynamically determined based on the mean time between failures of the inductor device and the actual operating conditions. The update cycle is determined in the early stages of equipment operation or when the operating conditions change frequently, and in the stable operation stage of the equipment. The transfer learning method is used to use the parameters of the original model as the initialization parameters of the new model. Only the newly collected data is trained to improve training efficiency. Step S5.2: Dynamically adjust the warning threshold and the classification standard according to the actual operation status and historical warning records. Use the Bayesian statistical method to combine the newly collected data and prior knowledge to update the probability distribution of the deviation degree, thereby adjusting the dynamic threshold range.

9. The intelligent failure trend prediction method for an inductor device according to claim 1, wherein It also includes data storage and management steps: Step S6, data storage and management, storing the collected raw data, pre-processed data, trend prediction results, deviation calculation results and warning information in a database using a distributed storage architecture; At the same time, a data indexing and query mechanism is established to facilitate the rapid retrieval and analysis of historical data, classify and store the data, and archive them according to the model, batch and production date information of the inductor components.

10. The intelligent failure trend prediction method for an inductor device according to claim 1, characterized in that: This method graphically displays the real-time status parameters, trend prediction curves, deviation and warning level information of the inductor device. The interface includes a real-time data display area, a trend chart area, a warning information prompt area and an operation control area. The change trend and prediction results of each parameter are intuitively presented through visual charts, which allows operators to quickly understand the operating status of the equipment.

Images