IoT-based hardware full-life-cycle quality traceability method and system

Abstract

The application relates to the technical field of hardware equipment quality tracing, and discloses a hardware full-life-cycle quality tracing method and system based on IoT, which comprises the following steps: acquiring multi-dimensional real-time data flow and arranging the same into a real-time running data set; extracting distribution characteristics based on a historical data archive to obtain a benchmark reference space; projecting the real-time data to calculate a deviation quantitative value, correcting environmental deviation when the deviation quantitative value exceeds a threshold value to obtain a deviation result; extracting degradation characteristics to match historical trajectories to locate a degradation process; performing time sequence analysis to predict residual life to obtain a failure time window; and generating and arranging maintenance intervention instructions into a scheduling instruction set. The method realizes hardware full-life-cycle quality tracing and early abnormality early warning, improves maintenance accuracy, and reduces fault loss.

Description

Technical Field

[0001] This invention relates to the field of hardware device quality traceability technology, and in particular to a hardware full lifecycle quality traceability method and system based on IoT. Background Technology

[0002] Currently, as hardware devices become increasingly complex in industrial production and daily life, quality traceability throughout their entire lifecycle has become a key link in ensuring reliability. How to combine IoT chips to test and maintain hardware has become a hot topic in the field.

[0003] In one existing technology, a combination of regular manual inspections or simple threshold alarm mechanisms is used to determine whether there are any abnormalities in the equipment by comparing real-time data with preset standard values, and then maintenance intervention can be carried out. However, in the face of dynamic changes in the equipment operating environment and the multidimensional distribution characteristics brought about by the differences in hardware devices, it is difficult to conduct in-depth analysis of massive amounts of collected data, filter out the impact of environmental interference on the data, and effectively explore the gradual degradation patterns contained in the data, resulting in deviations in the measurement of the deviation between real-time data and historical normal patterns.

[0004] Therefore, existing technologies suffer from the problem of unreliable extrapolation of hardware remaining lifespan through in-depth data analysis, leading to unreasonable timing of maintenance interventions. Summary of the Invention

[0005] This invention provides a hardware lifecycle quality traceability method and system based on IoT, to solve the problem in the prior art that it is difficult to reliably extrapolate the remaining lifespan of hardware through in-depth data analysis, resulting in unreasonable maintenance intervention timing.

[0006] Firstly, in order to solve the above-mentioned technical problems, the present invention provides a hardware lifecycle quality traceability method based on IoT, comprising: Acquire multidimensional real-time data streams during the operation of hardware devices, classify and organize the multidimensional real-time data streams to obtain a real-time running dataset; Based on a pre-set historical data archive, the multidimensional distribution characteristics of the real-time running dataset are extracted, and the sample distribution range of the multidimensional distribution characteristics is analyzed to obtain a benchmark reference space. The real-time running dataset is projected onto the reference space, and the distance between the real-time data points and the reference center is calculated to obtain the deviation metric value. The net deviation value is obtained by analyzing the deviation quantification. If the net deviation value exceeds the preset anomaly judgment threshold, the environmental impact deviation of the real-time data point is corrected by combining historical environmental fluctuation records to obtain the deviation result. Based on the deviation results, progressive degradation features are extracted. Based on the progressive degradation features, historical degradation trajectories with trend consistency that meet the preset trend consistency threshold are matched from the preset historical data archive to obtain degradation process location nodes. Based on the location nodes of the degradation process, the remaining lifespan of the equipment is predicted through time series analysis, and the equipment failure time window is obtained. Based on the equipment failure time window, maintenance intervention instructions are generated, and these instructions are organized to obtain a maintenance scheduling instruction set.

[0007] Secondly, the present invention provides a hardware lifecycle quality traceability system based on IoT, comprising: The data acquisition module is used to acquire multi-dimensional real-time data streams during the operation of hardware devices, classify and organize the multi-dimensional real-time data streams, and obtain real-time running datasets. The benchmark construction module is used to extract the multidimensional distribution features of the real-time running dataset based on a preset historical data archive, analyze the sample distribution range of the multidimensional distribution features, and obtain the benchmark reference space. The deviation calculation module is used to project the real-time running dataset onto the reference space, calculate the distance between the real-time data points and the reference center, and obtain the deviation quantification value. The environmental correction module is used to analyze the deviation quantification value to obtain the net deviation value. If the net deviation value exceeds the preset anomaly judgment threshold, the environmental impact deviation of the real-time data point is corrected by combining historical environmental fluctuation records to obtain the deviation result. The trajectory matching module is used to extract progressive degradation features based on the deviation results, and to match historical degradation trajectories with trend consistency that meet the preset trend consistency threshold from the preset historical data archive based on the progressive degradation features, so as to obtain the degradation process location node. The lifespan prediction module is used to locate nodes based on the degradation process, predict the remaining lifespan of the equipment through time series analysis, and obtain the equipment failure time window. The maintenance scheduling module is used to generate maintenance intervention instructions based on the equipment failure time window, organize the maintenance intervention instructions, and obtain a maintenance scheduling instruction set.

[0008] Compared with the prior art, the present invention has the following beneficial effects: (1) This invention solves the problem that the traditional fixed threshold method is difficult to adapt to individual differences of equipment and fluctuations in normal working conditions by constructing a benchmark reference space based on dynamic learning of historical data and calculating the deviation quantification value of real-time data. It realizes the leap from qualitative alarm to quantitative and accurate assessment of hardware health status, and significantly improves the sensitivity and accuracy of status monitoring.

[0009] (2) By introducing an environmental correction step, this invention actively retrieves and quantifies the impact of environmental fluctuations in the deviation analysis, and uses a regression model to separate them from the total deviation, effectively distinguishing between equipment degradation and environmental interference, fundamentally reducing the risk of false alarms and missed alarms caused by changes in environmental factors, and enhancing the robustness and reliability of the system in complex industrial environments.

[0010] (3) By matching similar historical degradation trajectories and locating degradation process nodes, this invention places the current short-term trend of the equipment in a complete and known fault evolution framework for comparison and extended prediction, replacing the limitation of traditional models that rely solely on recent data for mathematical extrapolation. This makes the prediction of the remaining lifespan and failure time window of the equipment more historically based and physically interpretable, and greatly improves the credibility of predictive maintenance.

[0011] (4) This invention generates a set of maintenance scheduling instructions that can be directly executed by multi-dimensional linkage and intelligent decision-making between the predicted failure time window and spare parts inventory, personnel scheduling and production plan. It realizes a closed loop from fault prediction to precise allocation of maintenance resources and seamless coordination with production plan, solves the problem of disconnect between prediction and execution, optimizes maintenance costs and ensures production continuity while ensuring equipment safety. Attached Figure Description

[0012] Figure 1 This is a schematic diagram of a hardware lifecycle quality traceability method based on IoT provided in the first embodiment of the present invention; Figure 2 This is a schematic diagram of a hardware lifecycle quality traceability system based on IoT provided in the second embodiment of the present invention. Detailed Implementation

[0013] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0014] Reference Figure 1 The first embodiment of the present invention provides a hardware lifecycle quality traceability method based on IoT, including the following steps: S11, acquire multi-dimensional real-time data streams during the operation of hardware devices, classify and organize the multi-dimensional real-time data streams to obtain real-time running datasets; S12, Based on the preset historical data archive, extract the multidimensional distribution characteristics of the real-time running dataset, analyze the sample distribution range of the multidimensional distribution characteristics, and obtain the benchmark reference space; S13, Project the real-time running dataset onto the reference space, calculate the distance between the real-time data points and the reference center, and obtain the deviation metric value; S14, Analyze the deviation quantification value to obtain the net deviation value. If the net deviation value exceeds the preset anomaly judgment threshold, combine the historical environmental fluctuation records to correct the environmental impact deviation of the real-time data points and obtain the deviation result. S15, Based on the deviation result, extract progressive degradation features, and based on the progressive degradation features, match historical degradation trajectories with trend consistency that meet the preset trend consistency threshold from the preset historical data archive to obtain degradation process location nodes. S16, locate the nodes according to the degradation process, predict the remaining lifespan of the equipment through time series analysis, and obtain the equipment failure time window; S17. Based on the equipment failure time window, generate maintenance intervention instructions, organize the maintenance intervention instructions, and obtain a maintenance scheduling instruction set.

[0015] In step S11, a multi-dimensional real-time data stream during the operation of the hardware device is acquired, and the multi-dimensional real-time data stream is classified and organized to obtain a real-time running dataset, including: The system acquires operational data during the operation of hardware devices, timestamps the operational data, and obtains a multidimensional real-time data stream. The fluctuation amplitude is calculated based on the multidimensional real-time data stream. If the fluctuation amplitude exceeds the preset fluctuation threshold, the corresponding time period of the multidimensional real-time data stream is marked to obtain the fluctuation sample. The fluctuation samples are frequency domain transformed to extract key features and construct a feature vector matrix; Cluster analysis is performed on the feature vector matrix to identify different fluctuation patterns and assign unique identifiers to obtain category identifiers; The data stream is distributed to the corresponding storage partition based on the category identifier, completing the categorized storage and obtaining the real-time running dataset.

[0016] It should be noted that the operational data encompasses physical parameters such as vibration, temperature, and current during equipment operation. Timestamp accuracy must be at the millisecond level to ensure multi-dimensional data time-series alignment and avoid misjudgments due to time discrepancies. The preset fluctuation threshold is dynamically adjusted based on the equipment's historical normal operating baseline, not a fixed value, accurately capturing potential early signs of faults. Frequency domain transformation employs Fast Fourier Transform (FFT), extracting key features including low-frequency energy (1-10Hz), mid-frequency peak values ​​(100-500Hz), and spectral entropy. Spectral entropy, calculated by the energy distribution dispersion of the frequency domain signal, quantifies signal complexity; a higher value indicates greater signal instability and reflects latent faults better than time-domain features. Cluster analysis uses the K-means algorithm, with the number of clusters adaptively determined based on historical data using the elbow rule, typically falling within 3-5 classes to balance classification granularity and computational efficiency. Class identifiers are used for intelligent data routing, reducing storage pressure and improving fault data retrieval efficiency.

[0017] It is worth noting that the historical normal operating baseline of the equipment requires continuous collection of fault-free operating data for at least 3 months, with a cumulative operating time of no less than 1000 hours. Statistical analysis is used to determine the distribution range and fluctuation patterns of each parameter, including core indicators such as the parameter mean, standard deviation, and maximum allowable fluctuation range. Normal operation refers to an operating state where all performance parameters of the equipment meet the factory technical specifications, there are no abnormal alarms or functional degradation, and no subsequent related failures occur. Abnormal operation refers to an operating state where parameters exceed the distribution range defined by the baseline, or although within the range, they show a continuous deterioration trend, potentially leading to a failure. The preset fluctuation threshold is dynamically adjusted based on this historical normal operating baseline. Initially, three times the standard deviation of each parameter in the baseline is used as the initial threshold. Subsequently, on a fixed monthly basis, newly added fault-free operating data is incorporated into the baseline for updates, the standard deviation of each parameter is recalculated, and the thresholds are adjusted synchronously. If the equipment operating conditions change, such as load adjustments or significant changes in the operating environment, a partial baseline update is immediately triggered, synchronously correcting the fluctuation thresholds of the corresponding parameters.

[0018] Faults can be categorized into latent and overt faults based on their manifestation. Latent faults are those where equipment has shown initial deterioration but does not exhibit obvious fault symptoms. They neither shut down nor show any apparent abnormalities, but rather exhibit subtle changes in the frequency domain characteristics of operating parameters. These subtle changes are specifically defined as: low-frequency energy fluctuations of more than 5% but not more than 15% relative to the baseline in the 1-10Hz range; peak harmonic components of mid-frequency components of 100-500Hz increasing by more than 3dB but not more than 8dB; or spectral entropy values ​​increasing by more than 0.05 but not more than 0.15 compared to the baseline. Overt faults, on the other hand, are those where equipment exhibits obvious abnormalities, such as shutdown, functional failure, or parameter mutations exceeding safe limits.

[0019] For example, taking a vehicle motor as the application, data is collected simultaneously using vibration, temperature, and current sensors, generating 1000 sampling points per second. Each data record is accompanied by a timestamp accurate to the millisecond. The historical normal operating baseline of the motor is constructed based on 1200 hours of fault-free operation data accumulated over four consecutive months since its manufacture. The mean vibration amplitude is 0.3 mm / s, the standard deviation is 0.1 mm / s, and the initial fluctuation threshold is set at 0.6 mm / s. Subsequent monthly updates to the baseline include newly added fault-free data. If the standard deviation of vibration amplitude in newly added data drops to 0.08 mm / s in a given month, the threshold is adjusted accordingly to 0.54 mm / s. During normal operation, the vibration amplitude remains stable within the range of 0.1-0.5 mm / s, conforming to the baseline distribution. When a sudden increase in vibration amplitude above 2.0 mm / s is detected during a certain period, exceeding the dynamically adjusted threshold, the system immediately marks the data from that period as a fluctuation sample. After performing a Fast Fourier Transform on the fluctuation samples, 10-20 dimensional frequency domain features are extracted to construct a feature vector matrix. K-means clustering is then used to classify them into three categories: slight imbalance, high-frequency harmonics, and impact, each assigned a unique identifier C1, C2, and C3, respectively. Subsequently, data in category C1 enters the regular monitoring database, while high-risk data in category C3 is directly pushed to high-priority storage and the alarm queue, completing the classification and storage to obtain the real-time running dataset.

[0020] In step S12, multidimensional distribution features of the real-time running dataset are extracted based on a preset historical data archive, and the sample distribution intervals of the multidimensional distribution features are analyzed to obtain a benchmark reference space, including: Key parameters are extracted from the real-time running dataset to form a multi-dimensional state vector; The multidimensional state vector is extracted using a preset sliding window, and the mean, standard deviation, and peak value of the multidimensional state vector within each time segment are calculated to construct a statistical feature matrix. The statistical feature matrix is ​​input into a preset historical data archive for matching and retrieval, and the historical normal operation samples with the smallest Euclidean distance are selected to obtain a benchmark comparison set; Calculate the covariance matrix of the benchmark comparison set, extract the principal component directions and project the sample points to determine the sample distribution interval; The sample distribution interval covering the projection position of the statistical feature matrix is ​​defined as the benchmark reference space.

[0021] It should be noted that the multidimensional state vector is the core representation of the equipment's operating status. Its dimensions correspond one-to-one with the types of core monitoring parameters of the equipment. For example, a vehicle motor can extract vibration amplitude, temperature value, and current intensity to form a three-dimensional vector. The sliding window length can be set to 5 seconds with a step size of 1 second, which can fully capture the short-term operating trends and fluctuation characteristics of the equipment, avoiding information lag caused by an excessively long window, while also preventing feature fragmentation caused by an excessively short window. The historical data archive needs to store normal operating samples of the equipment for more than one year, with a sample size of no less than 10,000. These samples need to cover the common normal operating conditions of the equipment and be able to truly reflect the multidimensional distribution patterns of normal equipment operation.

[0022] It is worth noting that core monitoring parameters are physical quantities or operating parameters that directly reflect the operating status, performance changes, and potential failure risks of key components of equipment. The monitoring parameters for different types of equipment are set according to their working principles and fault-sensitive points. For example, the core monitoring parameters for vehicle motors include vibration amplitude, operating temperature, and operating current, forming a three-dimensional state vector; the core monitoring parameters for industrial pumps include outlet pressure, motor speed, bearing temperature, and leakage, forming a four-dimensional state vector; and the core monitoring parameters for CNC machine tools include spindle vibration, cutting temperature, feed rate, and tool wear, forming a four-dimensional state vector.

[0023] Principal component directions are determined by eigenvalue decomposition of the covariance matrix, yielding multiple eigenvalues ​​and corresponding eigenvectors. The magnitude of each eigenvalue directly represents the information intensity carried by its corresponding eigenvector. Principal component directions are the directions pointed to by the eigenvectors with the highest eigenvalue ranking; these directions represent the areas with the greatest data variation and the most concentrated information. Selecting principal component directions with a cumulative variance contribution rate of at least 85% aims to reduce data dimensionality and simplify computational complexity while preserving the maximum amount of effective information from the original data, avoiding the loss of key features, and ensuring the reliability of subsequent analysis.

[0024] The sample distribution interval needs to cover the projection position of the current statistical feature matrix. The specific implementation process consists of four steps. First, project all sample points in the benchmark set and the currently constructed statistical feature matrix onto the extracted principal component directions to obtain their respective projection values. Second, for each principal component direction, calculate the maximum and minimum values ​​of the projection values ​​in the benchmark set; the range between these two values ​​is the sample distribution interval for that principal component direction. Third, check the projection values ​​of the current statistical feature matrix in each principal component direction to confirm whether they all fall within the sample distribution interval of the corresponding principal component. Fourth, if all projection values ​​are within the interval, it indicates that the device operating status corresponding to the current statistical feature matrix is ​​consistent with the distribution characteristics of historically operating samples; this multidimensional sample distribution interval can then be defined as the benchmark reference space. If there are cases where the projected values ​​exceed the range, the screening range of the benchmark comparison set is expanded. For example, the number of historical samples with the smallest Euclidean distance is increased from 100 to 200. The covariance matrix is ​​recalculated, the principal component direction and sample distribution range are extracted, until all projection positions of the current statistical feature matrix are covered, ensuring that the benchmark reference space can accurately adapt to the current normal operation status definition requirements of the device.

[0025] For example, core monitoring parameters of the vehicle motor, namely vibration amplitude, operating temperature, and operating current, are extracted from the real-time operating dataset to form a three-dimensional state vector. A 5-second sliding window is used to extract continuous data segments, with each window containing 5000 sampling points. The mean, standard deviation, and peak value of vibration amplitude, operating temperature, and operating current within each window are calculated, constructing a statistical feature matrix of 9 columns and 500 rows, denoted as matrix M_current. This matrix is ​​then compared for similarity with samples in a historical data archive. Each historical sample in the archive is also a feature vector composed of 9 features: mean, standard deviation, and peak value of vibration amplitude; mean, standard deviation, and peak value of operating temperature; and mean, standard deviation, and peak value of operating current, denoted as vector V_hist. To select the historical sample most similar to the current state, the similarity between the current matrix M_current and each historical sample vector V_hist needs to be calculated. Specifically, the Euclidean distance is calculated between each row of the matrix M_current (i.e., the feature vector of each time window) and the historical sample vector V_hist, resulting in 500 distance values. The average of these 500 distance values ​​is then used as the comprehensive distance metric between the matrix M_current and the historical sample V_hist. Finally, the 100 historical samples with the smallest comprehensive distance are selected to form the benchmark comparison set used for subsequent construction of the benchmark reference space.

[0026] In step S13, the real-time running dataset is projected onto the reference space, and the distance between the real-time data points and the reference center is calculated to obtain the deviation metric, including: The multidimensional state vector in the real-time running dataset is mapped to the reference space and a vector transformation is performed to obtain the projected coordinate vector; The absolute displacement is obtained by calculating the Euclidean distance between the projected coordinate vector and the geometric center point of the reference space. The absolute displacement is standardized using the variance of the reference space to eliminate the influence of dimensions and obtain the relative displacement. If the relative displacement exceeds a preset displacement threshold, the weighted Mahalanobis distance algorithm is used to correct the relative displacement to obtain a deviation quantification value.

[0027] It should be noted that the dimension of the high-dimensional state vector is equivalent to the number of core monitoring parameters of the device. For example, if a vehicle motor has three core monitoring parameters, the corresponding state vector is three-dimensional. The principal component directions are the eigenvector directions extracted in step S12, and they must meet the requirement that the cumulative variance contribution rate is not less than 85%. For example, if two principal component directions meet the requirements, a projection matrix is ​​constructed based on these principal component directions. The number of rows in the projection matrix equals the number of principal component directions, and the number of columns equals the dimension of the high-dimensional state vector. Each row of the matrix corresponds to an eigenvector of a principal component direction. Finally, the high-dimensional state vector is multiplied by the projection matrix to obtain the low-dimensional projected coordinate vector.

[0028] The geometric center point is the mean of the projected values ​​of all reference samples in the reference space, and is a vector formed by combining the mean values ​​of the projected values ​​of all reference samples in each principal component direction. The Euclidean distance is calculated based on the difference between the projected coordinate vector and the geometric center point in each dimension, obtained by taking the square root of the sum of the squares of the differences in each dimension.

[0029] The preset displacement threshold is set based on the fact that the projected values ​​of historical normal samples within the reference space follow an approximately normal distribution. Statistically, about 93.3% of the samples in a normal distribution fall within the range of the mean plus or minus 1.5 standard deviations. Setting the preset displacement threshold to 1.5 can cover parameter fluctuations under most normal operating conditions, avoiding missed detection of early faults. Furthermore, this value has been verified through extensive field testing on numerous industrial equipment, achieving accurate anomaly identification in quality traceability scenarios for various hardware devices such as motors, pumps, and machine tools, thus becoming the standard setting. If the application scenario requires higher early warning sensitivity, such as precision instruments, the threshold can be lowered to 1.2; if higher operational stability is required, such as heavy machinery, the threshold can be raised to 1.8.

[0030] It is worth noting that the modified formula for the weighted Mahalanobis distance algorithm is as follows: ; Where D is the corrected deviation metric, and n is the number of principal component directions. Let be the weight of the i-th principal component, which corresponds to the variance contribution rate of the principal component, and the sum of the weights of all principal components is 1. Let be the projection value of the real-time data point along the direction of the i-th principal component. This is the mean of the projection values ​​of all reference samples in the reference space along the direction of the i-th principal component, which is the value of the dimension corresponding to the geometric center point. Let be the standard deviation of the projection values ​​of all benchmark samples in the reference space along the i-th principal component direction. The principal component direction with the higher variance contribution rate carries more critical equipment status information, and its corresponding weight ratio is also larger.

[0031] Standardization can be achieved using the Z-score standardization method.

[0032] For example, all numerical calculations in the example are rounded to one decimal place to ensure the results are concise and conform to engineering display conventions. Taking a vehicle motor as the application object, its core monitoring parameters include vibration amplitude, operating temperature, and operating current, with a corresponding three-dimensional state vector of [0.3 mm / s, 45℃, 15A]. The process of mapping this vector to the reference space and calculating the deviation quantification is as follows: in step S12, two principal component directions have been extracted, with a cumulative variance contribution rate of 88%. The first principal component focuses on the combined influence of vibration amplitude and operating current, while the second principal component focuses on the influence of operating temperature.

[0033] First, using the statistics of the historical benchmark sample set used in step S12 to construct the benchmark reference space, the current three-dimensional state vector [0.3mm / s, 45℃, 15A] is Z-score standardized. In the historical benchmark sample set, the mean of vibration amplitude is 0.2mm / s and the standard deviation is 0.1mm / s; the mean of operating temperature is 44.0℃ and the standard deviation is 3.0℃; and the mean of operating current is 16.0A and the standard deviation is 2.0A. The standardized calculation is as follows: the standardized value of vibration amplitude is (0.3-0.2) / 0.1=1.0, the standardized value of operating temperature is 0.333, and the standardized value of operating current is -0.5, resulting in the standardized state vector [1.0,0.333,-0.5]. This is then multiplied by the projection matrix [0.80, 0.02, 0.04; 0.01, 0.98, 0.01]. The first principal component projection value is calculated as 0×0.80+0×0.02+0×0.04=0. Subsequently, the standardized state vector [1.0,0.333,-0.5] is multiplied by the 2×3 dimensional projection matrix, resulting in the first principal component projection value as 1.0×0.80 +0.333×0.02 + (-0.5)×0.04 =0.78666, which is rounded to one decimal place to 0.8. The projection value of the second principal component is 0.33134, which is rounded to one decimal place to 0.3. The final two-dimensional projected coordinate vector is [0.8, 0.3].

[0034] The geometric center point of the reference space is [0.1, 0.2] (this point is calculated based on the standardized projection value of historical reference samples). Substituting the corrected projected coordinate vector [0.8, 0.3], the difference in the first dimension is 0.8 - 0.1 = 0.7, the difference in the second dimension is 0.3 - 0.2 = 0.1, the sum of squares of the differences is 0.50, and the absolute displacement is approximately 0.71. In step S12, the standard deviation of the distribution of the first principal component was determined to be 0.8, the standard deviation of the distribution of the second principal component was determined to be 0.6, the combined standard deviation was 1.0, and the relative displacement was 0.71 / 1.0 = 0.71.

[0035] The preset displacement threshold is 1.5. The current relative displacement of 0.71 is less than the preset threshold, so there is no need to perform weighted Mahalanobis distance correction. The relative displacement of 0.71 is directly used as the deviation quantification value.

[0036] In step S14, the net deviation value is obtained by analyzing the quantified deviation value. If the net deviation value exceeds a preset anomaly judgment threshold, the environmental impact deviation of the real-time data points is corrected by combining historical environmental fluctuation records to obtain the deviation result, including: Based on the timestamp of the deviation from the quantified value, multidimensional factor data of the environment during the same period are retrieved to obtain the environmental fluctuation record sequence. The environmental fluctuation record sequence is processed using a preset time window to extract environmental change trends and construct local environmental feature vectors. The local environmental feature vector is input into the support vector regression model, and the quantitative value of the impact of the environment on the equipment status is output to obtain the baseline environmental impact deviation value. The difference between the quantified deviation value and the baseline environmental impact deviation value is calculated to obtain the net deviation value; If the net deviation value is greater than the preset anomaly judgment threshold, the environmental impact deviation of the real-time data points is corrected by combining the environmental fluctuation record sequence to obtain the deviation result.

[0037] It should be noted that the environmental multidimensional factor data includes ambient temperature, humidity, air pressure, external vibration, etc. The sampling frequency of these data must be completely consistent with the sampling frequency of the equipment operation data. For example, if the equipment operation data is collected at a frequency of 1000 sampling points per second, the environmental multidimensional factor data must also be collected synchronously at a frequency of 1000 sampling points per second. The preset time window and the sliding window length in step S12 are the same, both being 5 seconds.

[0038] The specific steps for training a support vector regression model are as follows: First, collect historical operating data of the device over the past 6 months, simultaneously extracting environmental multidimensional factor data and corresponding device deviation metrics during this period. The sample size should be no less than 5000 data points. Next, process the sample data, including performing Min-Max normalization on the environmental factor data and deviation metrics. Then, divide the preprocessed sample dataset into a training set and a test set in a 7:3 ratio. The training set is used for model parameter learning, and the test set is used for model performance validation. Next, select the radial basis function as the kernel function, and optimize the model's penalty and kernel parameters using a grid search method. Substitute these parameters into the training set for model training. Finally, use the test set to verify the model's prediction accuracy. If the prediction error is below a preset threshold, the model training is complete; if the error exceeds the threshold, additional samples or parameter adjustments are needed for retraining until the accuracy requirements are met.

[0039] The preset threshold here is set based on the standard accuracy requirements for quantifying the environmental impact of industrial equipment. It is determined by referencing the error range of support vector regression models for similar equipment. The core requirement is that the error must not affect the accuracy of the calculated environmental impact deviation value, avoiding misjudgments of the net deviation value due to model prediction bias. Based on extensive verification with real-world data, the preset threshold is set to 0.01. This value ensures that the deviation between the model's output quantified environmental impact value and the actual value is within an acceptable range and will not interfere with subsequent anomaly detection procedures.

[0040] An anomaly detection threshold is preset by first collecting all previously recorded fault cases from the equipment and extracting the net deviation value at the time of each fault. Next, the distribution of these net deviation values ​​is statistically analyzed, identifying the minimum value and the 95th percentile, with the minimum value used as the initial reference value for the threshold. Then, combined with the net deviation statistics during normal equipment operation, the false positive rate and false negative rate corresponding to the initial reference value are calculated. The false positive rate is the probability that a normal state is judged as an anomaly, and the false negative rate is the probability that a fault state is not judged. The threshold is then adjusted according to actual application needs. To reduce the false positive rate, the threshold can be appropriately increased; to reduce the false negative rate, the threshold can be appropriately decreased. After iterative adjustments, an optimal value balancing the false positive and false negative rates is determined and set as the threshold.

[0041] The threshold is typically set to 1.0, based on historical data from 200 similar industrial devices (including vehicle motors, industrial pumps, etc.). The 90th percentile of the net deviation value under normal operating conditions is 0.95, and this is adjusted to 1.0 for safety redundancy. In practical applications, calibration is required based on the device type.

[0042] For example, in the application scenario of a vehicle motor, the deviation quantification value calculated in step S13 is 1.64, corresponding to a timestamp of 14:30:00. Based on this timestamp, environmental multi-dimensional factor data from 14:30:00 to 14:30:05 are retrieved. Environmental sensors deployed around the motor synchronously collect data at a frequency of 1000 sampling points per second, collecting a total of 5000 sets of environmental data within 5 seconds. The data processing method is to take the arithmetic mean of the 1000 sampling points per second, obtaining one mean data point every second, and obtaining a total of 5 mean data points within 5 seconds, forming an environmental fluctuation record sequence. The specific data are as follows: ambient temperature 28℃, 28.3℃, 28.7℃, 29.1℃, 29.2℃, humidity 60%, 59.5%, 58.8%, 58.2%, 58.0%, and external vibration disturbance 0.5Hz, 0.49Hz, 0.51Hz, 0.48Hz, 0.52Hz. These continuous data together form an environmental fluctuation record sequence. The sequence was processed using a 5-second time window, and the core change indicators of each environmental parameter within the window were calculated: the rate of temperature rise was (29.2℃-28℃) / 5s=0.24℃ / s; the humidity fluctuation amplitude was characterized by the standard deviation of the humidity data within the 5-second window, and the standard deviation was calculated to be 0.82%, reflecting the intensity of humidity fluctuation; the mean of external vibration was (0.5+0.49+0.51+0.48+0.52) / 5=0.5Hz. Based on these three core indicators, a local environmental feature vector [0.24℃ / s, 3.33%, 0.5Hz] was constructed, and each feature was normalized using Min-Max to eliminate dimensional differences before being input into the support vector regression model.

[0043] The support vector regression model used was specifically trained for the motor of this vehicle model. The training data covered the historical operating data of the motor over the past 8 months, collecting a total of 8,500 valid samples, meeting the requirement of a minimum of 5,000 samples. The samples covered the full operating condition range of -10℃ to 45℃ ambient temperature, 30% to 80% ambient humidity, and 0.1 to 1.0Hz external vibration. After optimizing the penalty parameters and kernel parameters using a grid search method, the mean squared error of the test set was 0.008, which is lower than the preset threshold of 0.01. When the constructed local environmental feature vectors were input into the model, the model output a baseline environmental influence deviation value of 0.32.

[0044] This value represents the equipment deviation caused solely by environmental fluctuations under current environmental conditions. Its calculation is based on the model's learned correlation between the temperature rise rate (0.24℃ / s), humidity fluctuation amplitude (3.33%), average external vibration (0.5Hz), and equipment deviation. The difference between the quantified deviation value of 1.64 obtained in step S13 and the baseline environmental impact deviation value of 0.32 is calculated, yielding a net deviation value of 1.43. Referring to the historical fault data statistics report for this vehicle model's motor, its net deviation value is consistently below 0.9 during normal operation. The preset anomaly judgment threshold of 1.0 is the optimal value after calibration using 100 sets of fault cases. Since 1.43 is greater than 1.0, it indicates that the motor exhibits a genuine anomaly after excluding environmental influences. This net deviation value is the deviation result subsequently used for deterioration trajectory matching.

[0045] In step S15, based on the deviation result, progressive degradation features are extracted. Based on the progressive degradation features, historical degradation trajectories with trend consistency that meet a preset trend consistency threshold are matched from a preset historical data archive to obtain degradation process location nodes, including: The deviation results are processed by wavelet transform to obtain the trend component; Calculate the slope change rate and curvature characteristic value of the trend component, and integrate them to form a deterioration characteristic vector; Based on the degradation feature vector, historical degradation trajectories that meet the preset trend consistency threshold are retrieved from the preset historical data archive, and similar degradation trajectories are obtained after filtering and integration. Key time nodes that match the degradation feature vector in the similar degradation trajectory are extracted and mapped to normalized time coordinates to obtain the degradation process location nodes.

[0046] It should be noted that the wavelet transform uses a Daubechies 4th-order wavelet for a 3-level decomposition to extract the low-frequency trend component from the deviation signal. This component is a slowly changing, long-period part of the signal, corresponding to the gradual process of equipment degradation, and can effectively filter high-frequency noise such as sensor errors and transient electromagnetic interference. Wavelet transform is a well-known signal processing technique in this field, and its decomposition process can be implemented with reference to standard algorithms, such as the Mallat algorithm.

[0047] The number of decomposition levels is determined based on the frequency band separation requirements. By calculating the energy entropy ratio of each decomposition level, the minimum level (usually 3 levels) is selected to ensure that the energy concentration of the low-frequency components reaches more than 85%. If the number of levels is less than 3, the frequency band separation is insufficient, leaving residual mid-frequency interference (energy entropy ratio > 15%); if it is more than 3 levels, the low-frequency components are excessively smoothed (variance attenuation rate > 90%), affecting the sensitivity of degradation features. The Daubechies 4th order wavelet achieves a balance between smoothness and oscillation, making it suitable for industrial signal processing.

[0048] The rate of change of slope is obtained by calculating the ratio of the difference in slope of the trend components within adjacent time intervals to the time interval. An increase in the rate of change of slope indicates accelerated degradation, while a decrease indicates a slower rate of degradation. The curvature eigenvalue is calculated for discrete trend component sequences using the three-point circular arc method formula. The calculation formula is: ; In the formula, , , The time coordinates are three consecutive time points. , , These are the trend component values ​​at the corresponding time points. As the midpoint The corresponding curvature value. The average curvature is calculated by taking the arithmetic mean of the curvature values ​​obtained from all three consecutive points in the trend component sequence, reflecting the overall curvature degree of the degradation trend curve. The curvature feature value reflects the change in degradation mode; an increase in curvature means that the degradation mechanism has changed, such as from slight wear to severe wear. The two combined form the degradation feature vector.

[0049] The preset trend consistency threshold is set to 0.8, calculated using cosine similarity. This value was determined based on extensive industrial practice and data statistics. The cosine similarity value ranges from 0 to 1; the closer the value is to 1, the stronger the trend consistency between the two trajectories. In practice, if the threshold is below 0.8, a large number of trajectories with significant trend differences will be included, such as misjudging bearing wear trajectories and gear aging trajectories as similar, leading to subsequent degradation location errors. If the threshold is above 0.8, such as 0.9, over-selection will occur, resulting in too few trajectories that meet the criteria for degradation location. The threshold of 0.8 has been verified to ensure that the selected trajectories have a trend consistency of over 80% with the current degradation feature vector, and therefore has become the standard setting.

[0050] It is worth noting that the normalized time coordinate ranges from 0 to 1. 0 represents the initial deterioration state, that is, the stage when the equipment has just shown signs of deterioration but has not yet affected normal operation; 1 represents the failure state, that is, the stage when the equipment has reached the fault threshold and cannot work normally.

[0051] For example, taking a vehicle motor as the application object, the deviation result calculated in the early stage is a time series of 10 consecutive days, with one mean recorded for each day. The data are 1.37, 1.42, 1.49, 1.58, 1.69, 1.82, 1.97, 2.15, 2.36, and 2.60, respectively. A Daubechies fourth-order, third-level wavelet transform is performed on this deviation result. After the third-level decomposition, the low-frequency trend component is finally obtained, with the data being 1.38, 1.41, 1.48, 1.59, 1.70, 1.81, 1.96, 2.14, 2.35, and 2.59, respectively. This component is smooth and shows a continuous upward trend, consistent with the gradual deterioration law of bearing wear. When calculating the slope change rate, a time interval of 3 days is used. The slope of the first interval, i.e., the 1st to 3rd day, is (1.48-1.38) / 2=0.05. The slope of the third interval, i.e., the 7th to 9th day, is (2.35-1.96) / 2=0.195. The slope change rate is (0.195-0.05) / (9-3)=0.145 / 6≈0.02. The curvature feature value is calculated using the three-point circular arc method formula. The time coordinates of three consecutive points in the trend component sequence are taken as 1, 2, and 3, and the corresponding trend component values ​​are 1.38, 1.41, and 1.48. The curvature value of the intermediate point is obtained by substituting into the formula. After traversing all three consecutive points to calculate the curvature value, the arithmetic mean is taken to obtain the average curvature of the low-frequency trend component, which is 0.05. The two-dimensional deterioration feature vector [0.02, 0.05] is integrated. The preset historical data archive stores 120 degradation trajectory samples related to vehicle motors, covering various degradation types such as bearing wear, gear aging, and coil overheating.

[0052] By calculating the cosine similarity between the current degradation feature vector and the feature vector of each historical trajectory, 15 trajectories with a similarity of no less than 0.8 were selected, such as trajectories numbered 32, 47, and 61, with similarities of 0.83, 0.87, and 0.82, respectively. The dynamic time warping algorithm was used to calculate the alignment path cost between these 15 trajectories and the current degradation trend. The specific calculation steps are as follows: First, a distance matrix is ​​constructed, where the matrix elements are the Euclidean distances between corresponding points in the current degradation trend sequence and the historical degradation trajectory sequence. Second, a dynamic programming method is used to find the optimal path from the upper left corner to the lower right corner of the matrix, with the path constraint being that movement can only be to the right, down, or down to the right. Third, all distance elements on the optimal path are summed to obtain the initial path cost. Fourth, the initial path cost is normalized using a path length normalization method, dividing the initial path cost by the total number of points on the optimal path (i.e., the total number of steps in the path mapping) to obtain the cost per unit length, thereby eliminating the influence of sequence length differences on similarity evaluation. The final path cost is then obtained. The path cost is a dimensionless value; the smaller the value, the higher the alignment between the two trajectories.

[0053] Among them, trajectory number 47, marked as the vehicle motor bearing wear and deterioration trajectory, has the lowest path cost of only 0.03, and is therefore identified as the target similar trajectory. This target trajectory records the complete process of a certain model motor from the initial deterioration to failure, with a total deterioration time of 100 days. Key nodes include the slight wear start point (day 10, normalized coordinate 0.1), the wear acceleration point (day 35, normalized coordinate 0.35), the severe wear point (day 70, normalized coordinate 0.7), and the failure point (day 100, normalized coordinate 1.0). Comparing the current deterioration feature vector [0.02, 0.05] with the feature vector of the target trajectory, it is found that it perfectly matches the feature vector of the wear acceleration point (day 35). Therefore, it is mapped to a normalized time coordinate of 0.35, meaning that the current equipment is in the 35th stage of the deterioration process, corresponding to day 35 of the target trajectory, and there is still a 65-day safe operation buffer period.

[0054] In step S16, based on the location nodes of the degradation process, the remaining lifespan of the equipment is predicted through time series analysis to obtain the equipment failure time window, including: Based on the degradation process location node, extract the degradation trend data segment after the degradation process location node in the similar degradation trajectory; Calculate the first-order difference sequence of the degradation trend data segment, input it into a preset autoregressive model, generate difference prediction values, accumulate the difference prediction values ​​and restore them to the original data scale to obtain the degradation trend value; Compare the degradation trend value with the preset equipment failure threshold, and extrapolate the future degradation trend according to the timeline. If the degradation trend value is less than the preset equipment failure threshold, continue to check the next time point. If the degradation trend value is equal to or greater than the preset equipment failure threshold, record the theoretical failure time point. The confidence interval of the theoretical failure time point is calculated based on the preset autoregressive model. The theoretical failure time point is set as the center, and the confidence interval is set as the fluctuation range to obtain the equipment failure time window.

[0055] It should be noted that the autoregressive model adopts the ARMA(3,2) model, and its order is determined by optimization using the AIC criterion. The specific process is as follows: the order of the ARMA model is jointly determined by the autoregressive order p and the moving average order q, denoted as ARMA(p,q). First, the candidate range of the order is set based on the time span and sampling frequency of the equipment degradation trend data. Considering that the time dependence of industrial equipment degradation data is usually not too long, the candidate range of p is set to 1 to 5, and the candidate range of q is set to 1 to 4.

[0056] Subsequently, based on the extracted historical degradation trend data segments, an ARMA model was constructed for each (p,q) combination within the candidate range, and the AIC value of each model was calculated. The core of the AIC criterion is to balance the model fitting accuracy and complexity. Its calculation formula is AIC=2k-2ln(L), where k is the number of model parameters and L is the likelihood function value of the model. The more parameters there are, the higher the model complexity and the larger the value of k; the higher the fitting accuracy, the larger the likelihood function value L and the larger the value of ln(L). Finally, the model combination with the smallest AIC value is selected. The actual calculation results for vehicle motor degradation data show that when p=3 and q=2, the AIC value is -128.6, which is the smallest among all candidate combinations. For example, the AIC value of ARMA(2,2) is -125.3, and the AIC value of ARMA(3,3) is -126.1. Therefore, the optimal order is determined to be ARMA(3,2).

[0057] The core function of first-order differencing is to transform non-stationary degradation trends into stationary sequences. First, it's necessary to define a non-stationary degradation trend: a sequence whose statistical characteristics change over time, primarily manifested as non-constant mean, variance, or covariance. Taking equipment degradation as an example, the vibration amplitude degradation data of a vehicle motor due to bearing wear is a typical non-stationary sequence. As wear deepens, the vibration amplitude shows a continuous upward trend, with the mean gradually increasing from an initial 1.0 mm / s to 5.0 mm / s before failure. Simultaneously, the variance also increases over time. First-order differencing transforms this into a stationary sequence. The specific logic involves selecting the degradation values ​​of two adjacent time points within the degradation trend data segment, subtracting the value of the previous time point from the value of the latter, and obtaining the difference as a data point in the differencing sequence. This process is repeated throughout the entire degradation trend data segment to obtain the first-order differencing sequence. For example, the data on the deterioration of vehicle motor vibration amplitude are 1.52 mm / s (day 35), 1.61 mm / s (day 36), and 1.73 mm / s (day 37). The corresponding first-order difference sequences are 0.09 mm / s (1.61-1.52) and 0.12 mm / s (1.73-1.61). The mean of the transformed difference sequence is stable at around 0.1, and the variance fluctuation is extremely small, which meets the requirements of a stationary sequence.

[0058] The preset equipment failure threshold is determined through dual verification using both the equipment's factory standard and historical failure data. The specific steps are as follows: First, retrieve the critical parameter standard from the equipment's factory technical manual. This standard is formulated by the equipment manufacturer based on design limits, material properties, and safety specifications. For example, the factory critical value for the vibration amplitude of a certain model vehicle motor is 5.5 mm / s. Second, collect 120 sets of historical failure data for this model of motor. The statistical analysis shows that the distribution range of these values ​​is 4.8 mm / s to 5.6 mm / s, with 95% of the failure values ​​concentrated between 4.9 mm / s and 5.4 mm / s. The process involves three steps: First, the 95th percentile of historical failure values ​​is calculated to be 5.3 mm / s. Combined with the factory critical value of 5.5 mm / s, and considering the safety redundancy requirements in industrial applications, if the threshold is too close to the factory critical value, sudden failures may occur due to parameter fluctuations. If the threshold is too low, over-maintenance will result. Therefore, a balance between the two is chosen. Second, small-batch empirical testing is conducted to adjust the threshold. The operating data of 30 motors of the same model are tracked to verify that when the threshold is set to 5.0 mm / s, it can accurately identify the precursors of failure while allowing sufficient maintenance preparation time. Finally, this value is determined to be the preset equipment failure threshold.

[0059] It is worth noting that the confidence interval is calculated based on the model's prediction error variance, using a 95% confidence level. The core reason for choosing a 95% confidence level stems from a dual consideration of statistical regularities and industrial practice. From a statistical perspective, the prediction error of equipment degradation trends follows an approximately normal distribution. Under a normal distribution, approximately 95% of the sample data will fall within the range of "mean ± 1.96 times the standard deviation". This confidence level ensures that the predicted failure time point has a 95% probability of falling within the calculated time window.

[0060] The specific steps for calculating the confidence interval are as follows: First, obtain the prediction error variance through the training process of the ARMA(3,2) model. This variance is the variance statistic of the residuals after the model has fitted the historical data. For the vehicle motor degradation prediction scenario, the calculated prediction error variance is 1.78, and the corresponding standard deviation is... The first step is to calculate the marginal error as approximately 1.33, based on the statistical coefficient of 1.96 corresponding to the 95% confidence level. The second step is to calculate the marginal error as 1.96 × 1.33 ≈ 2.61, taking the time unit as days and rounding it to the nearest integer as 3 days. The third step is to add or subtract the marginal error to obtain the confidence interval, i.e. the fluctuation range of the failure time window, with the theoretical failure time point as the center.

[0061] For example, in the application scenario of vehicle motors, step S15 has determined that the normalized time coordinate of the degradation process location node is 0.55. The target similar trajectory corresponding to this node comes from the historical fault archive of a certain type of vehicle motor, with the archive number M-2024-047. This archive clearly records the complete degradation process of a certain type of motor of the same model in 2024 due to bearing wear, with a total degradation time of 100 days. Through normalized coordinate calculation, it can be found that the location node corresponds to the 35th day in the trajectory, specifically calculated as 0.35 multiplied by 100 days. This node is marked as a wear acceleration point in the target trajectory, and its corresponding bearing wear state transitions from slight wear to moderate wear, which perfectly matches the current degradation characteristics of the motor.

[0062] Based on the positioning node, the degradation trend data segment after day 35 is extracted from the similar trajectory of the target. This data segment represents the vibration amplitude degradation data from day 35 to day 100, sourced from the motor's operation and maintenance monitoring system, system number MMS-289. All data have undergone standardization and verification to ensure accuracy, with specific values ​​in mm / s, namely 1.52, 1.61, 1.73, 1.87, 2.03, 2.21, 2.42, 2.65, 2.91, 3.20, 3.53, 3.89, 4.28, 4.71, and 5.0. Value 5.0 represents the vibration amplitude value at day 100 when the motor fails due to severe bearing wear, perfectly matching the preset equipment failure threshold. Next, a first-order difference transformation is performed to calculate the difference between adjacent values, generating a difference sequence. The specific difference data calculations are as follows: 1.61-1.52=0.09, 1.73-1.61=0.12, ..., 5.0-4.71=0.29. After ADF stationarity testing, the mean of this difference sequence is 0.25, the variance is 0.008, and the test statistic is -4.26, which is less than the critical value of -3.43 at the 1% significance level, meeting the requirements for a stationary sequence and suitable for model input. This first-order difference sequence is then input into a pre-trained ARMA(3,2) model, specifically trained for vehicle motor bearing wear degradation scenarios. The training data comes from 80 historical degradation difference sequences of the same model of motor bearing wear, covering all stages of light, moderate, and severe wear. After the model's order is determined through AIC criterion optimization, it is validated on the test set, with a goodness-of-fit R-value. 2 The accuracy is 0.92, and the mean squared error is 0.007, which meets the accuracy standard for industrial equipment degradation prediction. The model generates differential prediction values ​​for the next 65 days, with some key prediction values ​​being 0.31 for day 1, 0.45 for day 10, 0.53 for day 20, ..., 0.82 for day 60, and 0.85 for day 65. Day 1 corresponds to day 36 of the trajectory, and day 65 corresponds to day 100 of the trajectory.

[0063] These differential prediction values ​​were then accumulated to the degradation value of 1.52 mm / s corresponding to the positioning node. This value represents the motor vibration amplitude on day 35. By accumulating and restoring the original data scale, the degradation trend value for the next 65 days was obtained. The key time points were 4.9 mm / s on day 64 and 5.0 mm / s on day 65. Day 64 corresponds to day 99 of the trajectory, calculated by accumulating the differential prediction values ​​from day 1 to day 64 sequentially by adding 1.52. Day 65 corresponds to day 100 of the trajectory, and the result exactly reached the preset equipment failure threshold of 5.0 mm / s. The basis for determining the preset equipment failure threshold of 5.0 mm / s is clear. It was verified by empirical data from 120 sets of historical failure data of the same model of motor, referencing the 95th percentile of historical failure vibration amplitude of 5.3 mm / s, and combining it with the critical vibration amplitude value of 5.5 mm / s specified in the equipment manufacturer's technical manual, fully considering the safety redundancy requirements in industrial applications.

[0064] The degradation trend value of 5.0 mm / s on day 65 has reached the threshold, therefore the theoretical failure time point is recorded. Combined with the current actual inspection date of March 10, 2025, which is also the execution date of steps S13 to S15, the relevant inspection records are fully stored in the maintenance log, log number LOG-20250310-02. Based on the current inspection date of March 10, 2025, and the predicted remaining lifespan of 65 days, the theoretical failure time point is May 14, 2025.

[0065] The confidence interval was then calculated. The variance of the prediction error of the ARMA(3,2) model for the bearing wear degradation trend was 1.78. This value is the variance statistical result of the residuals after fitting the model to historical data during the model training process. The corresponding standard deviation is... The calculated value is approximately 1.33. Based on the statistical coefficient of 1.96 corresponding to the 95% confidence level, the marginal error is calculated as 1.96 multiplied by 1.33, resulting in approximately 2.61. Considering that the time unit is days, the marginal error is rounded to an integer of 3 days, therefore the confidence interval is ±3 days. The final equipment failure time window is determined to be from May 11, 2025 to May 17, 2025. This time window has been entered into the vehicle motor's operation and maintenance management system, system number OMS-756. Based on this time window, the maintenance team plans to complete the bearing repair and replacement work before May 8, 2025.

[0066] In step S17, maintenance intervention instructions are generated based on the equipment failure time window, and these instructions are organized to obtain a maintenance scheduling instruction set, including: Calculate the time difference between the current system time and the device failure time window to obtain the safe operation buffer period; Based on the safe operation buffer period, retrieve spare parts inventory and maintenance personnel hour schedule, assess maintenance feasibility, and generate maintenance intervention instructions; The earliest intervention time in the maintenance intervention instructions is analyzed, compared with the production plan deadline, and the maintenance information is integrated to obtain a maintenance scheduling instruction set.

[0067] It's important to note that the core definition of the safe operation buffer period is the time interval between the current system time and the starting boundary of the equipment failure time window. This is the core basis for developing maintenance plans. Its primary function is to allow sufficient preparation and execution time for maintenance work, preventing sudden equipment failure before maintenance is completed. The specific calculation logic for the safe operation buffer period is the difference between the starting boundary time of the equipment failure time window and the current system time. The starting boundary of the failure time window is chosen as the calculation benchmark because it represents the earliest possible time point when equipment failure occurs. The buffer period calculated based on this is the "shortest safe buffer duration," which minimizes risk. Calculating based on the middle or end boundary of the window might lead to maintenance delays due to an overestimated buffer period, or premature equipment degradation causing failures. A consistent time unit must be used in the calculation, typically days. If the calculation involves months or years, the number of days in each calendar month must be precisely accumulated.

[0068] The construction of the spare parts inventory database and maintenance personnel work hour schedule includes the following steps: The spare parts inventory database needs to record the available quantity and inventory location of key spare parts, and synchronously update the spare parts dispatch cycle, which refers to the time from submitting a dispatch request to the spare parts being delivered to the maintenance site. During the assessment, first, the available quantity of the target spare parts, such as vehicle motor bearings, is retrieved to confirm whether it meets the minimum maintenance requirement, which is at least one set. Then, the corresponding dispatch cycle is queried based on the inventory location to determine whether the spare parts can be delivered within the safe operation buffer period. Subsequently, personnel suitability verification is performed. The maintenance personnel work hour schedule records in detail the skill level, certification qualifications, monthly work schedule, and idle time of each maintenance personnel. During the assessment, personnel with the target equipment maintenance qualifications, such as a vehicle motor maintenance special certificate, are first screened. Then, the idle time of the screened personnel within the safe operation buffer period is checked to confirm that there are continuous and sufficient working hours to complete the maintenance. Taking the replacement of motor bearings as an example, this maintenance requires 3 hours, so it is necessary to ensure that personnel have 3 or more consecutive hours of idle time.

[0069] Finally, a time matching verification is performed, comparing the estimated delivery time of spare parts with the earliest available time for maintenance personnel. If there is an overlapping time period where spare parts have been delivered and personnel are available, and this overlapping time period still has a redundancy of at least 2 days before the start boundary of the failure window (this redundancy is used to deal with unexpected maintenance issues), then maintenance is deemed feasible. If any level of verification fails, such as insufficient spare parts, lack of suitable personnel, or no overlapping time, then maintenance is deemed temporarily infeasible. In this case, an emergency procedure needs to be initiated, such as emergency procurement of spare parts or temporary allocation of personnel.

[0070] It is worth noting that the earliest intervention time and the production schedule deadline are key indicators for balancing maintenance and production, with the core purpose of avoiding maintenance work from interfering with the normal production schedule. The earliest intervention time refers to the earliest point in time when the conditions for maintenance work can be carried out, and it is determined by the later of the estimated delivery time of spare parts and the earliest available time of maintenance personnel.

[0071] The production plan deadline refers to the planned completion time of the production line or production task currently being performed by the equipment. Generated by the production management system, it clarifies the final delivery node of the production task and is directly related to production progress and capacity targets. If the earliest intervention time is earlier than the production plan deadline, it means that performing maintenance at this time will interrupt the ongoing production task, resulting in capacity loss. The maintenance time should be adjusted to the first overlapping idle period after the production plan deadline. If the earliest intervention time is later than or equal to the production plan deadline, maintenance can be performed according to the earliest intervention time without interfering with production.

[0072] The maintenance scheduling instruction set includes intervention time, spare parts list, maintenance personnel configuration, and priority tags. The priority tags are used to identify the urgency of maintenance tasks. Their core function is to help the operations and maintenance department quickly prioritize tasks, allocate resources to handle high-urgency and high-impact maintenance needs, and improve operations and maintenance efficiency.

[0073] Priority tags are primarily based on three core dimensions. The first is equipment importance; core production equipment, such as vehicle drive motors, has a higher priority than auxiliary equipment like cooling fans, as vehicle drive motors directly impact the entire production line. The second dimension is the length of the safe operation buffer period; the shorter the buffer period, the higher the urgency. The third dimension is the scope of the failure's impact; equipment that would cause production line shutdowns, safety hazards, or significant economic losses after failure has a higher priority than equipment affecting only a single process. The industry standard is a three-tier priority system. High priority applies to core equipment, buffer periods of no more than 15 days, or scenarios with significant impact from failure; these scenarios require maintenance to be initiated within 48 hours. Medium priority applies to important equipment, buffer periods between 15 and 30 days; these scenarios require maintenance to be initiated within 72 hours. Low priority applies to auxiliary equipment, buffer periods exceeding 30 days; these scenarios can be maintained according to regular schedules. Priority tags must be prominently displayed in the maintenance scheduling instruction set as the core basis for resource allocation.

[0074] For example, in the application scenario of vehicle motor bearing wear, the operation and maintenance management system is numbered OMS-756, the production management system is numbered PMS-345, and the spare parts inventory management system is numbered IMS-567. Step S16 has clearly defined the equipment failure time window as May 11, 2025 to May 17, 2025, and the current system time is March 10, 2025. Relevant records are stored in the operation and maintenance log, which is numbered LOG-20250310-03. First, the safe operation buffer period is calculated by subtracting the current system time March 10, 2025 from the failure time window start boundary May 11, 2025. Specifically, the calculation is broken down as follows: March 10 to March 31, a total of 22 days; April, a total of 30 days; and May 1 to May 10, a total of 10 days, for a total buffer period of 22 + 30 + 10 = 62 days. The buffer period is long enough to meet the time requirements for maintenance preparation and execution, laying the foundation for maintenance feasibility assessment.

[0075] A maintenance feasibility assessment was conducted based on a 61-day safe operation buffer period. The first step was to verify the availability of spare parts. The spare parts inventory management system was searched; the system number is IMS-567. The inventory record for the target spare part, the vehicle motor bearing, was checked. The bearing model is B-2024-01, consistent with the motor bearing model in example S16. The inventory record showed an available quantity of 8 sets, located in Spare Parts Warehouse No. 2, Area A, Row 3, with a transportation cycle of 1 day. This transportation cycle means that after submitting a transportation request, it can be delivered to the maintenance site within one working day. The spare parts quantity meets the maintenance requirements; only one set of spare parts is needed for maintenance. Furthermore, the short transportation cycle allows for rapid delivery within the buffer period. Therefore, the spare parts verification passed.

[0076] The second step was to verify personnel suitability. The maintenance personnel's work schedule was checked; this schedule, numbered SCH-202503, is updated by the Human Resources department before the 5th of each month. Personnel with specialized vehicle motor repair qualifications and a skill level of at least level 3 were screened. Three qualified personnel were ultimately selected: Engineer Li, Engineer Wang, and Engineer Zhang. Engineer Li's skill level is level 3 (certificate number W-2023-589); Engineer Wang's skill level is level 4; and Engineer Zhang's skill level is level 3. Their idle time during the 61-day buffer period was checked. It was found that Engineer Li had no scheduled work from April 20th to April 25th, 2025, representing a continuous period of idle time. Engineers Wang and Zhang were also handling other equipment repair tasks during this period and had no idle time. Engineer Li's idle time was 6 days, far exceeding the 3 hours required for motor bearing replacement; therefore, the personnel suitability verification passed.

[0077] The third step is to perform time matching verification. The spare parts dispatch cycle is 1 day. If the dispatch request is submitted on April 19th, the parts can be delivered to the maintenance site on April 20th. Engineer Li's earliest available time is April 20th, and the overlapping period is from April 20th to April 25th. This overlapping period still has 21 days of redundancy before the failure window starts on May 11th. Based on the requirement that the redundancy time must be no less than 2 days, 21 days is far above the threshold, so the verification passes. Therefore, maintenance is deemed feasible, and a maintenance intervention instruction is generated.

[0078] The core content of the maintenance intervention instruction is as follows: the recommended intervention time is April 20, 2025, which is the earliest intervention time. The required spare part is one set of vehicle motor-specific bearings, model B-2024-01. The assigned maintenance personnel is Engineer Li, whose skill level is level 3, and the initial priority is marked as medium. Further analysis of the earliest intervention time, April 20, 2025, compares it with the production plan deadline. The current production plan for the production line where the motor is located is retrieved from the production management system. The system number is PMS-345, and the production plan number is PRO-202504. The production plan shows that the production task is the processing of a certain batch of vehicle drive components, with a planned deadline of April 22, 2025. Since the earliest intervention time, April 20, 2025, is earlier than the production plan deadline of April 22, 2025, direct maintenance would affect the production schedule. Therefore, the maintenance intervention time is adjusted to the first idle period after the production plan deadline. This period is April 23, 2025. On April 23, Li Gong was still free and had no other production tasks arranged.

[0079] Finally, all information was integrated to form a complete maintenance scheduling instruction set. This instruction set, numbered MD-20250310-01, includes the following: 1. Intervention Time: April 23, 2025, 9:00-12:00. Three hours of maintenance time are reserved during this period. After maintenance, a 2-hour no-load test run and a 1-hour load test run are required for verification. 2. Spare Parts List: One set of vehicle motor-specific bearings, model B-2024-01, dispatch request number TR-20250419-01, required to be delivered to the maintenance site before April 20th. 3. Maintenance Personnel Configuration: Main maintenance personnel: Mr. Li, contact number 138XXXX5678; Assistant: Mr. Zhao, responsible for tool preparation and on-site coordination, skill level 2. 4. Priority Tag: Low Priority. The classification is based on the fact that although the equipment is a core drive motor, its safe operation buffer period is 62 days. Exceeding the 30-day threshold, it is classified as low priority according to the priority rules. If actual production needs necessitate an upgrade in priority, a manual review process must be initiated, and the reason for the upgrade must be noted in the instruction set. The buffer period is 61 days, and the impact of the failure is limited to the production process corresponding to a single motor, with no urgent safety hazards. 5. Supporting requirements: A shutdown application must be submitted to the production department one day prior to maintenance. The day before maintenance is April 22nd. After maintenance, a 2-hour no-load test run and a 1-hour load test run must be conducted. The test run data must be entered into the operation and maintenance management system for record-keeping, and then reviewed and accepted by the quality department.

[0080] In summary, this invention discloses a hardware lifecycle quality traceability method based on IoT. By combining IoT technology and integrating multi-dimensional degradation feature analysis, it achieves accurate quality traceability of hardware devices throughout their entire lifecycle. This solves problems such as difficulty in filtering environmental interference, unreliable lifespan prediction, and unreasonable maintenance timing in existing technologies, providing strong support for stable equipment operation.

[0081] Reference Figure 2 The second embodiment of the present invention provides a hardware lifecycle quality traceability system based on IoT, comprising: The data acquisition module is used to acquire multi-dimensional real-time data streams during the operation of hardware devices, classify and organize the multi-dimensional real-time data streams, and obtain real-time running datasets. The benchmark construction module is used to extract the multidimensional distribution features of the real-time running dataset based on a preset historical data archive, analyze the sample distribution range of the multidimensional distribution features, and obtain the benchmark reference space. The deviation calculation module is used to project the real-time running dataset onto the reference space, calculate the distance between the real-time data points and the reference center, and obtain the deviation quantification value. The environmental correction module is used to analyze the deviation quantification value to obtain the net deviation value. If the net deviation value exceeds the preset anomaly judgment threshold, the environmental impact deviation of the real-time data point is corrected by combining historical environmental fluctuation records to obtain the deviation result. The trajectory matching module is used to extract progressive degradation features based on the deviation results, and to match historical degradation trajectories with trend consistency that meet the preset trend consistency threshold from the preset historical data archive based on the progressive degradation features, so as to obtain the degradation process location node. The lifespan prediction module is used to locate nodes based on the degradation process, predict the remaining lifespan of the equipment through time series analysis, and obtain the equipment failure time window. The maintenance scheduling module is used to generate maintenance intervention instructions based on the equipment failure time window, organize the maintenance intervention instructions, and obtain a maintenance scheduling instruction set.

[0082] It should be noted that the IoT-based hardware lifecycle quality traceability system provided in this embodiment of the invention is used to execute all the process steps of the IoT-based hardware lifecycle quality traceability method in the above embodiment. The working principles and beneficial effects of the two are one-to-one, so they will not be described again.

[0083] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0084] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A method for hardware lifecycle quality traceability based on IoT, characterized in that, include: Acquire multidimensional real-time data streams during the operation of hardware devices, classify and organize the multidimensional real-time data streams to obtain a real-time running dataset; Based on a pre-set historical data archive, the multidimensional distribution characteristics of the real-time running dataset are extracted, and the sample distribution range of the multidimensional distribution characteristics is analyzed to obtain a benchmark reference space. The real-time running dataset is projected onto the reference space, and the distance between the real-time data points and the reference center is calculated to obtain the deviation metric value. The net deviation value is obtained by analyzing the deviation quantification. If the net deviation value exceeds the preset anomaly judgment threshold, the environmental impact deviation of the real-time data point is corrected by combining historical environmental fluctuation records to obtain the deviation result. Based on the deviation results, progressive degradation features are extracted. Based on the progressive degradation features, historical degradation trajectories with trend consistency that meet the preset trend consistency threshold are matched from the preset historical data archive to obtain degradation process location nodes. Based on the location nodes of the degradation process, the remaining lifespan of the equipment is predicted through time series analysis, and the equipment failure time window is obtained. Based on the equipment failure time window, maintenance intervention instructions are generated, and these instructions are organized to obtain a maintenance scheduling instruction set.

2. The method for hardware lifecycle quality traceability based on IoT according to claim 1, characterized in that, The process of acquiring multidimensional real-time data streams during the operation of the hardware device, classifying and organizing the multidimensional real-time data streams to obtain a real-time running dataset includes: The system acquires operational data during the operation of hardware devices, timestamps the operational data, and obtains a multidimensional real-time data stream. The fluctuation amplitude is calculated based on the multidimensional real-time data stream. If the fluctuation amplitude exceeds the preset fluctuation threshold, the corresponding time period of the multidimensional real-time data stream is marked to obtain the fluctuation sample. The fluctuation samples are frequency domain transformed to extract key features and construct a feature vector matrix; Cluster analysis is performed on the feature vector matrix to identify different fluctuation patterns and assign unique identifiers to obtain category identifiers; The data stream is distributed to the corresponding storage partition based on the category identifier, completing the categorized storage and obtaining the real-time running dataset.

3. The method for hardware lifecycle quality traceability based on IoT according to claim 1, characterized in that, The process involves extracting multidimensional distribution features from the real-time running dataset based on a pre-defined historical data archive, analyzing the sample distribution intervals of these multidimensional distribution features, and obtaining a benchmark reference space, including: Key parameters are extracted from the real-time running dataset to form a multi-dimensional state vector; The multidimensional state vector is extracted using a preset sliding window, and the mean, standard deviation, and peak value of the multidimensional state vector within each time segment are calculated to construct a statistical feature matrix. The statistical feature matrix is ​​input into a preset historical data archive for matching and retrieval, and the historical normal operation samples with the smallest Euclidean distance are selected to obtain a benchmark comparison set; Calculate the covariance matrix of the benchmark comparison set, extract the principal component directions and project the sample points to determine the sample distribution interval; The sample distribution interval covering the projection position of the statistical feature matrix is ​​defined as the benchmark reference space.

4. The method for hardware lifecycle quality traceability based on IoT according to claim 3, characterized in that, The step of projecting the real-time running dataset onto the reference space, calculating the distance between the real-time data points and the reference center, and obtaining the deviation metric includes: The multidimensional state vector in the real-time running dataset is mapped to the reference space and a vector transformation is performed to obtain the projected coordinate vector; The absolute displacement is obtained by calculating the Euclidean distance between the projected coordinate vector and the geometric center point of the reference space. The absolute displacement is standardized using the variance of the reference space to eliminate the influence of dimensions and obtain the relative displacement. If the relative displacement exceeds a preset displacement threshold, the weighted Mahalanobis distance algorithm is used to correct the relative displacement to obtain a deviation quantification value.

5. The method for hardware lifecycle quality traceability based on IoT according to claim 1, characterized in that, The analysis of the deviation quantification value yields a net deviation value. If the net deviation value exceeds a preset anomaly judgment threshold, the environmental impact deviation of the real-time data points is corrected by combining historical environmental fluctuation records to obtain the deviation result, including: Based on the timestamp of the deviation from the quantified value, multidimensional factor data of the environment during the same period are retrieved to obtain the environmental fluctuation record sequence. The environmental fluctuation record sequence is processed using a preset time window to extract environmental change trends and construct local environmental feature vectors. The local environmental feature vector is input into the support vector regression model, and the quantitative value of the impact of the environment on the equipment status is output to obtain the baseline environmental impact deviation value. The difference between the quantified deviation value and the baseline environmental impact deviation value is calculated to obtain the net deviation value; If the net deviation value is greater than the preset anomaly judgment threshold, the environmental impact deviation of the real-time data points is corrected by combining the environmental fluctuation record sequence to obtain the deviation result.

6. The method for hardware lifecycle quality traceability based on IoT according to claim 1, characterized in that, The step involves extracting progressive degradation features based on the deviation results, and then matching historical degradation trajectories with trend consistency that meet a preset trend consistency threshold from a preset historical data archive based on these features to obtain degradation process location nodes, including: The deviation results are processed by wavelet transform to obtain the trend component; Calculate the slope change rate and curvature characteristic value of the trend component, and integrate them to form a deterioration characteristic vector; Based on the degradation feature vector, historical degradation trajectories that meet the preset trend consistency threshold are retrieved from the preset historical data archive, and similar degradation trajectories are obtained after filtering and integration. Key time nodes that match the degradation feature vector in the similar degradation trajectory are extracted and mapped to normalized time coordinates to obtain the degradation process location nodes.

7. The method for hardware lifecycle quality traceability based on IoT according to claim 6, characterized in that, The step of locating nodes based on the degradation process, predicting the remaining lifespan of the equipment through time series analysis, and obtaining the equipment failure time window includes: Based on the degradation process location node, extract the degradation trend data segment after the degradation process location node in the similar degradation trajectory; Calculate the first-order difference sequence of the degradation trend data segment, input it into a preset autoregressive model, generate difference prediction values, accumulate the difference prediction values ​​and restore them to the original data scale to obtain the degradation trend value; Compare the degradation trend value with the preset equipment failure threshold, and extrapolate the future degradation trend according to the timeline. If the degradation trend value is less than the preset equipment failure threshold, continue to check the next time point. If the degradation trend value is equal to or greater than the preset equipment failure threshold, record the theoretical failure time point. The confidence interval of the theoretical failure time point is calculated based on the preset autoregressive model. The theoretical failure time point is set as the center, and the confidence interval is set as the fluctuation range to obtain the equipment failure time window.

8. The method for hardware lifecycle quality traceability based on IoT according to claim 1, characterized in that, The step involves generating maintenance intervention instructions based on the equipment failure time window, organizing these instructions, and obtaining a maintenance scheduling instruction set, including: Calculate the time difference between the current system time and the device failure time window to obtain the safe operation buffer period; Based on the safe operation buffer period, retrieve spare parts inventory and maintenance personnel hour schedule, assess maintenance feasibility, and generate maintenance intervention instructions; The earliest intervention time in the maintenance intervention instructions is analyzed, compared with the production plan deadline, and the maintenance information is integrated to obtain a maintenance scheduling instruction set.

9. A hardware lifecycle quality traceability system based on IoT, characterized in that, include: The data acquisition module is used to acquire multi-dimensional real-time data streams during the operation of hardware devices, classify and organize the multi-dimensional real-time data streams, and obtain real-time running datasets. The benchmark construction module is used to extract the multidimensional distribution features of the real-time running dataset based on a preset historical data archive, analyze the sample distribution range of the multidimensional distribution features, and obtain the benchmark reference space. The deviation calculation module is used to project the real-time running dataset onto the reference space, calculate the distance between the real-time data points and the reference center, and obtain the deviation quantification value. The environmental correction module is used to analyze the deviation quantification value to obtain the net deviation value. If the net deviation value exceeds the preset anomaly judgment threshold, the environmental impact deviation of the real-time data point is corrected by combining historical environmental fluctuation records to obtain the deviation result. The trajectory matching module is used to extract progressive degradation features based on the deviation results, and to match historical degradation trajectories with trend consistency that meet the preset trend consistency threshold from the preset historical data archive based on the progressive degradation features, so as to obtain the degradation process location node. The lifespan prediction module is used to locate nodes based on the degradation process, predict the remaining lifespan of the equipment through time series analysis, and obtain the equipment failure time window. The maintenance scheduling module is used to generate maintenance intervention instructions based on the equipment failure time window, organize the maintenance intervention instructions, and obtain a maintenance scheduling instruction set.

Images