Method for predictive maintenance of a device with anomaly prediction capabilities

Abstract

The application discloses a device predictive maintenance method with abnormal prediction capability, and relates to the technical field of device predictive maintenance, and comprises the following steps: constructing a direction continuous factor based on a direction change scalar and a historical statistical fluctuation value, using the direction continuous factor to control a numerical rescaling function in a normalization process, so that the normalization processing retains the trend characteristics of original data; converting the normalized data into a continuous curvature space, forming a trend potential by analyzing the local slope continuity of a data change path, and generating a trend energy expression vector based on the trend potential; performing multi-order positive and negative axis decomposition on the trend energy expression vector, converting each axis change into an independent distribution curve, and generating a trend deviation index through probability increment superposition convolution. The application solves the problem of difficult identification of weak trend degradation, and realizes early warning and accurate maintenance response of device state change.

Description

Technical Field

[0001] This invention relates to the field of predictive maintenance technology, and more specifically to a predictive maintenance method for equipment with anomaly prediction capabilities. Background Technology

[0002] Predictive maintenance, which involves predicting anomalies, refers to the use of multi-source operational data (such as temperature, vibration, current, voltage, and pressure) collected from various industrial equipment in industrial settings. By incorporating advanced algorithms such as machine learning, deep learning, or time-series modeling, a predictive model is built to identify potential operational anomalies in real time. This allows for the prediction of possible abnormal states before actual equipment failures occur, and the development or triggering of corresponding maintenance strategies based on the predictions. This approach differs from traditional periodic preventative maintenance and post-failure repair maintenance; it emphasizes "data-driven proactive perception," significantly reducing unplanned downtime and minimizing the waste of maintenance resources, thereby improving overall equipment availability and production continuity. In industrial environments, equipment systems are complex and operate under harsh conditions. Any single point of failure can cause entire production lines to shut down or even lead to safety accidents. Therefore, deploying predictive maintenance mechanisms with anomaly prediction capabilities helps achieve intelligent, refined, and efficient equipment operation and maintenance management, serving as a crucial support for industrial intelligent upgrading and cost reduction.

[0003] In industrial settings, existing predictive maintenance technologies for equipment with anomaly prediction capabilities typically achieve intelligent equipment maintenance through a collaborative process involving five key stages: data acquisition, data preprocessing, state modeling, anomaly detection and prediction, and maintenance decision-making. First, the system continuously collects multi-dimensional data during equipment operation using various sensors deployed on the equipment (such as vibration sensors, temperature sensors, and current / voltage detection devices). Then, the collected data undergoes preprocessing steps such as filtering, normalization, and missing data completion to ensure its usability for modeling and analysis. In the state modeling stage, the system introduces machine learning algorithms (such as random forests and support vector machines) or deep learning models (such as LSTM and GRU) to establish a dynamic predictive model of the equipment's operating state, learning in real time the boundary characteristics between normal and abnormal operation. Next, by performing trend prediction or deviation analysis on the data, the system can identify potential operational anomalies or deterioration trends, and even predict specific time windows when failures may occur. Finally, combining the anomaly prediction results with the equipment's importance level, the system automatically generates or recommends maintenance strategies, such as replacing components in advance, adjusting operating conditions, and scheduling planned maintenance, thereby avoiding losses from sudden failures. Overall, this technology system is data-driven, model-centric, and scenario-oriented, providing an efficient, controllable, and sustainable intelligent maintenance method for industrial equipment.

[0004] The existing technology has the following shortcomings:

[0005] In equipment operating status data, some key parameters may exhibit a continuous and stable unidirectional micro-acceleration trend within a local time window. Although the amplitude of this trend is extremely small, it often represents early signs of performance degradation within the equipment, such as load accumulation, micro-deformation of structural components, or loss of precision. In this case, because such trend signals do not show obvious abrupt changes in value, existing predictive maintenance technologies with anomaly prediction capabilities typically use fixed interval scaling or Z-score normalization methods during data normalization. This results in these low-amplitude gradual changes being compressed into approximately constant values, causing their true trend changes to be smoothed out or lost in the modeling input, making them unusable as effective features for the model to perceive and utilize. Because existing technologies do not consider the identification mechanism for such "low-amplitude trend-type" degradation features, they cannot accurately determine the trend-based deviation behavior of the equipment based on these status data trajectories. This causes the system to "blind" early degradation processes, rendering predictive maintenance methods incapable of early warning. Ultimately, this may lead to maintenance responses being triggered only when the equipment reaches a critical point, increasing the risk of failure and maintenance costs.

[0006] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0007] The purpose of this invention is to provide a predictive maintenance method for equipment with anomaly prediction capabilities, in order to solve the problems in the background art described above.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a predictive maintenance method for equipment with anomaly prediction capabilities, specifically comprising the following steps:

[0009] S1. Construct a multi-time-level data acquisition window, generate multiple non-overlapping time intervals for device status parameters, perform differential calculations within each time interval, and generate a directional change scalar based on the direction of time series change.

[0010] S2. Construct a direction continuity factor based on the directional change scalar and historical statistical fluctuation values. Use the direction continuity factor to control the numerical recalibration function in the normalization process so that the normalization process retains the trend characteristics of the original data.

[0011] S3. The normalized data is transformed into a continuous curvature space. By analyzing the local slope continuity of the data change path, trend potential energy is formed, and path integration is performed based on this trend potential energy to generate a trend energy expression vector.

[0012] S4. Decompose the trend energy expression vector into multi-order positive and negative axes, transform each axial change into an independent distribution curve, and generate the trend offset index by convolution with probability increments.

[0013] S5. Establish the mapping interval of the trend deviation index, construct an asymmetric distribution model using the critical change range in historical samples, set a three-segment state classification interval, and map and classify the current trend deviation index.

[0014] S6. Based on the continuous rate of change of the trend deviation index, the local structure of the scalar of the trend change is deduced, and the time level weights and normalization function parameters are dynamically updated through the difference inversion mechanism to realize the online correction of the trend determination parameters.

[0015] Preferably, S1 specifically includes:

[0016] A basic sampling period is set, and multiple sets of time windows of different lengths and non-overlapping time windows are constructed in the time dimension. Each time window is divided into multiple sampling points according to the basic sampling period, forming a data hierarchical structure with short-term, secondary-term and medium-to-long-term scales.

[0017] Within each time window, the difference between adjacent data is calculated point by point for the corresponding state parameter data sequence to generate a complete difference sequence, which is used to characterize the direction and magnitude of change within that time period.

[0018] Based on the statistical proportion of positive and negative directions of the difference sequence, and combined with the absolute average magnitude of the difference, a directional change scalar is generated to represent the directional consistency and trend strength of state changes within the time window.

[0019] Preferably, S2 specifically includes:

[0020] Historical status parameter data of the equipment under fault-free conditions are collected, and multiple sliding time windows are constructed based on a fixed sampling period. The maximum value, minimum value, mean and standard deviation within each sliding time window are calculated to form a set of historical statistical fluctuation values ​​for reference.

[0021] The ratio of the directional change scalar in the current time period to the corresponding historical standard deviation is calculated to obtain the directional continuity factor used to represent the trend direction and the relative change magnitude.

[0022] The maximum or minimum boundary of the normalization function is adjusted according to the positive or negative value of the direction continuity factor so that the normalized state parameter data retains the original change direction characteristics.

[0023] The system performs a consistency comparison of the change direction before and after normalization on the normalized state parameter data. When the consistency ratio is lower than the set threshold, it returns to adjust the continuity factor of the direction and re-normalizes.

[0024] Preferably, S3 specifically includes:

[0025] By treating the normalized state parameter sequence as a continuous change path, we analyze the change direction and bending intensity between each data segment in time sequence, and construct a curvature space structure with time traceability.

[0026] Based on the constructed curvature structure, the direction of data change is determined to be continuous in each continuous time segment to form a trend continuity label, and a weight corresponding to the trend direction is assigned to each trend segment.

[0027] Based on the curvature characteristics and trend continuity weights at each time point, a trend potential energy sequence is generated. The trend potential energy value is used to represent the trend direction and change magnitude of that point in the state evolution process.

[0028] The trend potential energy sequence is accumulated and integrated according to a preset time period to form a trend energy expression vector containing multiple dimensions, which is used for subsequent state trend analysis and classification identification.

[0029] Preferably, S4 specifically includes:

[0030] The trend energy expression vector is decomposed into positive sub-vectors and negative sub-vectors according to the numerical signs of each dimension, which correspond to the expression paths of the strengthening trend and weakening trend of the state parameters in each time interval, respectively.

[0031] Based on the positive and negative sub-vectors, trend probability density distribution curves are constructed to reflect the distribution of the probability of occurrence and the magnitude of trends in each direction over time.

[0032] The probability density curves of positive and negative trends are converted into trend increment functions, and weighted convolution is performed at the corresponding time positions to form a trend offset index, which is used to characterize the direction and intensity of net trend change.

[0033] Preferably, S5 specifically includes:

[0034] Collect historical operating data of the equipment, generate a trend deviation index for each historical time point, compare and label it with the status label corresponding to that time point, and build a correspondence between the trend deviation index and the equipment operating status.

[0035] Based on the aforementioned correspondence, the distribution characteristics of the trend deviation index under various states are statistically analyzed, an asymmetric distribution model is established, and the trend deviation index is divided into three state classification intervals: central stable interval, negative degradation interval, and positive enhancement interval.

[0036] The currently collected trend deviation index is mapped to the classification interval in real time. The equipment operating status is determined based on the mapping result, and predictive maintenance decisions are executed in conjunction with these decisions, including risk warnings, maintenance suggestion pushes, or monitoring frequency adjustment operations.

[0037] Preferably, S6 specifically includes:

[0038] Based on the continuous rate of change of the trend deviation index, anomaly windows of trend deviation are identified, and a deviation tracing mechanism is initialized to locate the time period in which potential trend expression is missing.

[0039] The scalar structure of the guidance change within the trend offset anomaly window is compared with the time-period difference, the ideal guidance structure for the corresponding time period is reconstructed, and a trend expression offset difference map is generated based on the structural deviation.

[0040] Based on the trend expression offset difference map, the weight allocation of the normalization function boundary parameters and the time level acquisition window is dynamically adjusted, and the parameter evolution process of each round of adjustment is recorded to achieve non-interrupted online correction of the trend determination parameters.

[0041] The technical effects and advantages provided by the present invention in the above technical solution are as follows:

[0042] 1. This invention effectively solves the problem in existing technologies of inaccurate identification of low-amplitude trend degradation, leading to the failure of early warning. By constructing a multi-time-level data acquisition structure and combining a directional change scalar with a direction continuity factor, the system not only achieves a structural expression of the temporal micro-variation trend of equipment operating parameters, but also ensures, through a dynamic recalibration mechanism of the normalization function, that the micro-amplitude but continuous trend signal is not compressed or disappeared during the normalization process, thus improving the identifiability of trend features in subsequent modeling inputs. Furthermore, by mapping the normalization results to a continuous curvature space and constructing trend potential and trend energy expression vectors, the system achieves dynamic modeling and structural quantification of the data evolution path, effectively enhancing the overall ability to characterize the trend direction, intensity, and persistence.

[0043] 2. This invention achieves accurate classification of equipment operating states at different trend evolution stages through the construction of a trend offset index and the state mapping mechanism of an asymmetric distribution model; it has significant advantages, especially in identifying mild degradation or slight structural imbalance trends. The trend offset index not only integrates directional, probability distribution characteristics, and incremental rate information, but also dynamically adjusts the normalization function parameters and time acquisition weights through the identification of trend offset anomaly windows and the guided expression inversion mechanism, realizing online model correction and continuous optimization of trend judgment accuracy. The overall solution possesses high trend sensitivity, model adaptability, and deployment stability, effectively supporting early risk detection, proactive maintenance decision-making, and optimized control of overall operating costs during long-term equipment operation, and has significant technical promotion value. Attached Figure Description

[0044] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0045] Figure 1 This is a flowchart illustrating the predictive maintenance method for equipment with anomaly prediction capabilities according to the present invention. Detailed Implementation

[0046] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, they are provided so that the description of this disclosure will be more complete and fully convey the concept of the exemplary embodiments to those skilled in the art.

[0047] This invention provides, for example Figure 1 The predictive maintenance method for equipment with anomaly prediction capabilities, as shown, specifically includes the following steps:

[0048] S1. Construct a multi-time-level data acquisition window, generate multiple non-overlapping time intervals for device status parameters, perform differential calculations within each time interval, and generate a directional change scalar based on the direction of time series change.

[0049] In this embodiment, to achieve high-precision identification of trend changes in device status parameters, especially unidirectional micro-acceleration trend changes that appear in the early stages with extremely low amplitude, a multi-time-level data acquisition window is constructed as follows, and a data-guided change scalar representing the trend direction is generated within each window. This process includes the following steps:

[0050] First, a basic sampling period is established, and a set of multi-time-level data acquisition windows with a defined time span is constructed. During operation, the device generates raw data streams of multiple state parameters in real time, such as temperature, current, rotational speed, and vibration signals. For these state parameters, a unified basic sampling period must first be defined as the smallest unit of analysis in the time domain, for example, setting data acquisition to occur once per second. Based on this period, multiple non-overlapping time windows of different lengths are constructed along the time dimension; for example, the first layer window has a length of 10 sampling points, the second layer has a length of 30 sampling points, and the third layer has a length of 60 sampling points. These windows are arranged sequentially on the timeline, forming a hierarchical set of windows with no temporal overlap. By setting windows in a hierarchical manner, the system can capture short-term, mid-term, and medium-to-long-term trend information at different time scales, providing sufficient time structure support for identifying subtle trends.

[0051] Secondly, within each data acquisition window, time-series differencing is performed on the state parameters within that window. Specifically, for the state data sequence within each window, starting from the first sampling point, the numerical difference between the current sampling point and its predecessor is calculated, and this process is repeated for the entire window. The difference between each pair of adjacent data points reflects the direction and magnitude of change within that time period. During the differencing process, the complete difference value sequence is preserved without compression or thresholding to ensure that any small but continuous trends are accurately recorded. This step provides raw gradient information for subsequent trend direction determination, ensuring that the trend formation process is not obscured by preprocessing.

[0052] Taking temperature parameters during equipment operation as an example, assuming the temperature data sequence collected within a certain non-overlapping time interval is: [65.01, 65.03, 65.06, 65.10, 65.15, 65.21], then according to the differential calculation method, the difference between adjacent sampling points is calculated sequentially, resulting in the differential sequence: [+0.02, +0.03, +0.04, +0.05, +0.06]. This sequence reflects a stable, unidirectional, and gradually increasing upward trend in temperature within this time interval. Although the change amplitude at each step is extremely small, the consistent direction and increasing amplitude of the difference precisely indicate the early signs of equipment load accumulation or gradual degradation of system heat dissipation performance. In this implementation step, this differential sequence will be completely preserved as the basic information for subsequent calculation of directional change scalars, ensuring that the system has the ability to identify such continuous micro-trends.

[0053] Third, directional features of the time series are extracted from each difference sequence, and a corresponding directional change scalar is generated. The extraction of directional features is based on the statistical analysis of the positive and negative directions of all difference terms within the window. The sign of each difference value is marked, the number of positive and negative differences is counted, and their proportional differences are calculated. Then, a directional consistency factor is constructed by combining the absolute average amplitude of the difference values ​​to measure whether the changes within the window exhibit continuous and skewed characteristics. Based on this, a specific directional change scalar is calculated by multiplying the directional consistency factor by the average difference amplitude to quantify the directional trend and stability of state changes within the window. This directional change scalar will serve as a key data feature for maintaining trend characteristics in the next stage of data normalization and trend identification.

[0054] Taking the difference sequence [+0.02, +0.03, +0.04, +0.05, +0.06] as an example, all differences in this sequence are positive, indicating that the state parameters show a continuous upward trend within this time window. When extracting directional features, the number of positive differences is first counted as 5, and the number of negative differences is 0. The directional proportion difference is calculated as (5−0) / (5+0) = 1.0, indicating extremely high directional consistency. Then, the absolute average amplitude of the difference values ​​is calculated as (0.02+0.03+0.04+0.05+0.06) / 5 = 0.04. Multiplying the directional proportion difference (1.0) by the average amplitude (0.04) yields a directional change scalar of 0.04. This value not only reflects the uniformity of the change direction (all positive) but also the relative strength of the trend (stable upward movement), and can be used for the next stage of normalization processing and trend maintenance judgment. It is a quantitative description of the slightly accelerated change trend within this time window.

[0055] Finally, the directional change scalars calculated for each non-overlapping time interval are used as scalar results representing the directionality of the trend within that interval, and stored in a time-series consistent data structure for use in the next step. No trend combination or inductive processing is performed on them. In this step, the final output of the data processing is a set of directional change scalars with a time-series order; it does not constitute a trend expression or trend sequence structure, which will be constructed in subsequent steps.

[0056] S2. Construct a direction continuity factor based on the directional change scalar and historical statistical fluctuation values. Use the direction continuity factor to control the numerical recalibration function in the normalization process so that the normalization process retains the trend characteristics of the original data.

[0057] In this embodiment, to address the technical problem of information compression and trend loss that traditional normalization methods easily cause when processing low-amplitude unidirectional trend changes, a method is proposed that constructs a directional continuity factor based on a directional change scalar and historical statistical fluctuation values, and uses this factor to control the numerical recalibration function in the normalization process, thereby achieving trend-preserving normalization processing. This embodiment specifically includes:

[0058] Calculate the historical statistical fluctuation set of the target state parameter: By selecting historical state parameter data collected over a long period under fault-free operation (e.g., data collected by a temperature sensor under stable operating conditions over the past week), a sliding time window is constructed with a fixed sampling period. Within each window, the maximum, minimum, mean, and standard deviation of the parameter are extracted. This results in a summary interval of statistical results from multiple windows, representing the natural fluctuation range of the state parameter under normal conditions. The statistical fluctuation set reflects the range of variation of the state parameter under non-abnormal operating conditions and will serve as a reference basis for subsequent trend comparison and normalized interval recalibration.

[0059] The directional continuity factor is calculated based on the scalar change in directional trend and historical statistical volatility values: The historical standard deviation obtained in the previous step is used as a benchmark for volatility intensity. This benchmark is then normalized and ratioed to the scalar change in directional trend calculated within the current time interval. This ratio is calculated by dividing the scalar change in directional trend by the corresponding standard deviation, yielding the trend strength ratio. This ratio measures the relative strength of the current trend within the background volatility. Furthermore, if the scalar change in directional trend is positive, the directional continuity factor is positive; if the scalar change in directional trend is negative, the directional continuity factor is negative. The final directional continuity factor is a signed real number, where its absolute value represents the significance of the trend, and its sign indicates the direction of the trend. This factor will be used as a control weight during the normalization process.

[0060] Taking the vibration acceleration parameter of the equipment as an example, assuming that the calculated scalar change in orientation is +0.015 within a certain time window, it indicates that the data in this segment shows a stable upward trend. Meanwhile, the standard deviation of this parameter, extracted based on the equipment's historical stable operating conditions, is 0.005. Then, according to the normalized ratio calculation, dividing the scalar change in orientation by the historical standard deviation yields a trend strength ratio of 0.015 ÷ 0.005 = 3.0. Since the scalar change in orientation is positive, it indicates that the current trend is positive; therefore, the direction continuity factor is +3.0. The positive sign of this direction continuity factor indicates that the trend direction is continuously strengthening, and the absolute value of 3.0 indicates that the strength of this trend is significantly enhanced within the historical background fluctuation range. In subsequent normalization processing, this direction continuity factor will be used as a dynamic scaling factor to adjust the upper limit of the normalization interval, thereby ensuring that this small but continuous trend change is not distorted or lost due to normalization compression, effectively preserving its trend characteristics for further model identification and learning.

[0061] The numerical rescaling function during the normalization process is controlled using a directional continuity factor. When normalizing the current state parameters, the commonly used linear scaling formula is (X−min) / (max−min), where min and max are statically set upper and lower boundaries. In this implementation, to prevent trend signals from being flattened during normalization, the boundary values ​​of max or min are dynamically adjusted based on the value of the directional continuity factor. If the directional continuity factor is positive, indicating a positive trend, max is increased to enhance the significance of upward changes; if the directional continuity factor is negative, min is decreased to preserve the subtle changes in negative trends. The boundary adjustment ratio is controlled by the absolute value of the directional continuity factor, and a reasonable limit is set to prevent amplification of abnormal fluctuations. This ensures that trend changes are not compressed into near-zero invalid values ​​during normalization due to boundary mismatch.

[0062] For example: Let's assume a certain state parameter has an initial observed value of 5.12 within the current time interval. The minimum value of this parameter in the historical statistical fluctuation range is 4.90, and the maximum value is 5.20. The default normalization interval is [4.90, 5.20]. Within this time interval, based on the directional change scalar obtained in the previous step (+0.015) and the historical standard deviation (0.005), the calculated directional continuity factor is +3.0, indicating a strong positive trend for this parameter during this period. To prevent this positive trend from being compressed into a weak signal close to constant during normalization due to an overly narrow normalization interval, the upper boundary of the normalization interval is dynamically adjusted based on the absolute value of the directional continuity factor. Multiplying the absolute value of the directional continuity factor by the proportional adjustment coefficient α = 0.005 yields an upper boundary correction of 0.015, resulting in a new normalization upper boundary of 5.215. The normalization formula is updated to: (X − 4.90) ÷ (5.215 − 4.90). The original value of 5.12 is now normalized to approximately 0.954, whereas without boundary adjustments, the normalized value would only be approximately 0.733. This method effectively enhances the normalized representation of the current trend signal, preventing it from being treated as an insignificant state and thus weakened in subsequent modeling.

[0063] To avoid excessive distortion of the normalization boundary due to abnormal extreme values ​​of the directional continuity factor, the system sets an upper limit threshold for the directional continuity factor when executing the recalibration function, for example, limiting its absolute value to no more than 5.0. When the directional continuity factor exceeds this upper limit, its cutoff value is used as the calculation basis to prevent the boundary from expanding too quickly and affecting the stability of the data distribution. In terms of boundary adjustment strategy, the updates of the normalization upper and lower limits adopt an asymmetric mechanism: if the directional continuity factor is positive, only the normalization upper limit is adjusted; if the directional continuity factor is negative, only the normalization lower limit is adjusted; if the directional continuity factor is close to zero, no boundary adjustment is performed. This processing logic can enhance the sensitivity of trend direction while avoiding overall value range shift. Furthermore, the system can introduce a sliding window mechanism to perform mean smoothing on the directional continuity factor over multiple time periods, generating a stable boundary correction factor, making the normalization boundary time-extended and robust, ensuring consistency in trend tracking across multiple time periods, thereby ensuring the continuity and discriminability of the trend throughout the entire prediction process.

[0064] The system outputs a sequence of state parameters after trend-preserving normalization and verifies the trend preservation effect: The state parameter data processed by the numerical rescaling function undergoes a directional consistency check, comparing the direction of data change (positive or negative) before and after normalization to ensure consistency. If the preservation rate exceeds a set threshold (e.g., 95%), the normalization process is considered to have not disrupted the original trend; if the direction of change reverses or the rate of change weakens sharply, the system returns to adjusting the direction continuity factor and re-executes the normalization process until the trend preservation performance meets the target. This verification mechanism ensures that trend preservation is not only at the method design level but also implemented in the perceptible results of the normalized data output, guaranteeing that subtle trends can continue to participate in the prediction process in subsequent modeling stages.

[0065] To ensure that the trend-preserving normalization process is not only theoretically feasible in terms of its methodological structure but also verifiable in actual output results, this implementation further introduces a quantifiable verification mechanism for orientation consistency testing. Specifically, the original state parameter sequence before normalization and the processed result after normalization are subjected to difference operations to form two sets of difference sequences. Subsequently, by comparing whether the change direction (i.e., the difference sign) of each pair of adjacent data points in the two difference sequences is consistent, the orientation consistency ratio is calculated, which is the ratio of the number of point pairs with consistent positive and negative signs to the total number of point pairs. If this ratio is greater than or equal to a set threshold (e.g., 95%), the normalization process is considered to have performed well in preserving the trend directionality, and the process can proceed to the subsequent trend modeling process; if the ratio is lower than the threshold, the system will automatically trigger the correction mechanism of the normalization process, locally re-evaluate the weight contribution of the direction continuity factor according to the time period in which the difference direction reversal point is located, adjust the recalibration boundary, and perform normalization processing again. Through this closed-loop verification and correction mechanism, this implementation method ensures that trend features are continuous, responsive, and perceptible throughout the entire data processing chain, thereby improving the system's ability to identify subtle trend-based degradation behaviors at an early stage.

[0066] S3. The normalized data is transformed into a continuous curvature space. By analyzing the local slope continuity of the data change path, trend potential energy is formed, and path integration is performed based on this trend potential energy to generate a trend energy expression vector.

[0067] In this embodiment, the normalized data is transformed into a continuous curvature space. By analyzing the local slope continuity of the data change path, a trend potential energy is formed. Based on this trend potential energy, path integration is performed to generate a trend energy expression vector, specifically including:

[0068] The normalized state parameter data is mapped into a curvature space structure with continuous expressive capabilities. After completing trend-preserving normalization, the system obtains a sequence of state parameter values ​​arranged at equal time intervals. To further identify the changing structure of the trend, this numerical sequence is first considered as a dynamic trajectory arranged in chronological order. By coherently analyzing the morphological changes of adjacent data points in this trajectory, a data structure space with curvature concepts is constructed. Specifically, within each local data interval formed by a set of continuous sampling points, the overall bending direction and bending intensity of the values ​​in that interval are identified, serving as the basic geometric characterization of the state change trend within that interval. Throughout the formation of the curvature structure of the entire data sequence, the chronological order is maintained, ensuring that each curvature change can be traced back to its original time point. This mapping operation allows the system to not only focus on the magnitude of changes at individual points when analyzing numerical trends but also to observe the continuous structure between adjacent data points, thus endowing the numerical sequence with an expressive ability that possesses direction awareness and structural characteristics. After completing this mapping, the system will transform the normalized state data into a set of continuous change paths with temporal morphological information, laying the foundation for in-depth calculations of subsequent trend direction and trend inertia.

[0069] In further implementation, to make the curvature space mapping process more reproducible in engineering, the system uses a sliding interval method to process the normalized state parameter data. Specifically, a fixed-length sliding window is set and slids sequentially from the beginning of the sequence, containing several consecutive equally spaced sampling points. Within each sliding window, the system observes the morphological changes of the local data segment, determining whether there are obvious upward arches, downward concavities, or directional reversals. For example, if the state values ​​within a certain sliding window are 0.712, 0.716, 0.721, 0.727, and 0.734 respectively, then this data segment shows a stable upward trend, and the value changes gradually, indicating that the trajectory has a consistent upward curvature structure. Based on this, the system marks it as a "positive curvature enhancement segment" in the curvature space. Conversely, if a data sequence shows a reversal in the middle, i.e., the value first rises and then falls, or vice versa, the system identifies it as a "bending and turning segment" and marks it as a "curvature reversal zone" in the curvature space. Through the above processing, the system not only constructs the overall trend change path of the data, but also clarifies the geometric structural features and temporal positions of the trend. This provides a spatiotemporally locatable structural basis for subsequent identification of the trend's inertial continuity, offset intensity, and abnormal breakthrough points, and realizes the expansion of trend cognition ability from one-dimensional numerical sequences to two-dimensional structural expression.

[0070] Based on the curvature spatial structure, the system identifies the slope variation trend of each segment in a continuous time path, forming a trend continuity evaluation mechanism. After obtaining a data path with curvature characteristics, the system performs a time-series sliding analysis to determine whether the data path's variation trend is consistent across continuous time periods. Specifically, within each continuous time segment, the system extracts the direction of change between adjacent data points and determines whether this direction remains continuous throughout the segment, i.e., whether the direction of change frequently reverses. If the data variation direction remains consistent across multiple consecutive time points, it can be considered that the data segment exhibits significant trend inertia. The system records these consecutive time periods of trend consistency and labels their directional attributes, such as continuous upward or continuous downward. Furthermore, the system can also assign different weight values ​​to each time period based on the stability of the trend direction to reflect the strength of the data segment's contribution to the overall trend. This weight not only represents the continuity of the trend but also expresses the significance of the trend within the numerical structure. The resulting trend continuity label will serve as one of the driving factors in the subsequent trend potential energy generation process, enabling the trend potential energy to include not only the magnitude of numerical changes but also the trend inertia strength reflected by directional consistency.

[0071] To ensure the feasibility and stability of the trend continuity evaluation mechanism, the system employs overlapping sliding intervals to analyze the entire time series segment by segment when processing the curvature space structure. Within each sliding interval, the system extracts the direction of change of all adjacent data points and calculates the consistency of these directions. For example, if the normalized state data within a certain sliding interval are 0.703, 0.707, 0.712, 0.718, and 0.725, then four sets of adjacent data differences can be extracted, all of which are positive, indicating a stable upward trend in this segment. Based on this, the system assigns the segment a "positive high consistency" label and further assigns a higher trend continuity weight to this trend segment, such as 0.9 or above. Conversely, if the direction of data change within a certain interval alternates frequently, such as 0.712, 0.708, 0.715, 0.710, and 0.718, the system identifies it as a "directionally unstable segment" and assigns it a lower weight, such as 0.2. Furthermore, the system can combine multiple dimensions such as the duration of the trend within each interval, the retention rate of the trend direction, and the degree of trend deflection to conduct a composite evaluation of each trend interval and establish a trend inertia strength level. For example, a segment with consistent direction for five consecutive time points and an increasing change in magnitude is rated as a "strong trend inertia segment" to drive subsequent potential energy construction; while a segment with consistent change for only two time points is marked as a "weak trend segment" to weaken its influence in subsequent processing. In this way, the system can construct a trend continuity label system that includes both direction and strength for each position in the trend path, so that the generation of trend potential energy not only depends on the morphological characteristics of the values, but also strengthens the identification of the consistency characteristics of the trend direction, effectively improving the responsiveness and expression accuracy of weak trends.

[0072] A trend potential energy expression sequence is constructed based on trend continuity and curvature characteristics to quantify the trend strength and direction at each position in the data path. The generation process of trend potential energy is based on the aforementioned continuous curvature expression and trend continuity identification. At each data point, the system simultaneously refers to the path curvature characteristics and trend continuity weight of that point, combining the two to evaluate whether the point has a significant trend. If a point has obvious path bending characteristics and the trend direction within its segment is highly consistent, it indicates that the point plays a trend-driving role in the current state sequence. The system assigns a higher trend potential energy value to such points; conversely, in areas where the trend direction frequently reverses or data changes are not significantly bent, a lower potential energy value is assigned, indicating that the trend significance is not significant. In the process of forming a complete trend potential energy expression sequence, the system always maintains the continuity of the time series structure, ensuring that each trend potential energy value corresponds to a specific time position. This trend potential energy expression sequence can be regarded as a trend evolution description with directionality, structure and dynamism, which transforms the original state parameters from a single numerical sequence into a trend expression object with "behavioral attributes", laying the foundation for further integrating trend potential energy and forming a discriminable feature vector.

[0073] To further ensure the stability and interpretability of the trend potential energy expression sequence, the system employs a combined scoring strategy for each time point when constructing the trend potential energy. This strategy uses the path curvature characteristics and trend continuity weights obtained in previous steps as input factors for joint evaluation. For example, if the data change curve at a given time point exhibits a clear upward arching trend and falls within a consistently upward range across six consecutive time points, the system determines that this point possesses significant trend-driving attributes in the current sequence and assigns it a high trend potential energy value, such as 0.87. Conversely, if the curvature at a data point is approximately zero or lacks a continuous direction, and the direction of change within its time period fluctuates frequently, the system considers this point a non-trend-dominant point and assigns it only a low trend potential energy value, such as 0.12. The entire trend potential energy expression sequence is composed of the trend potential energy values ​​at all time points, maintaining a one-to-one correspondence with the original state parameter sequence, thus ensuring complete temporal continuity. In practice, the trend potential energy sequence can be visualized as a trend strength curve that fluctuates over time, enabling equipment maintenance personnel to clearly identify which time points or time segments constitute areas of trend enhancement, trend stability, or trend weakening. Through this method, the system successfully transforms a single numerical change trajectory into a structured trend expression with behavioral descriptive attributes, providing a clear, rich, and time-series-traceable trend signal foundation for subsequent trend aggregation, offset index generation, and health status identification.

[0074] The system transforms the trend potential energy expression sequence into a trend energy expression vector that represents the entire trend and outputs it for subsequent trend analysis and classification. After constructing the trend potential energy sequence, the system performs an overall structural summary of the sequence, extracting a set of vector features reflecting the overall trend behavior. Specifically, the system integrates the trend potential energy expression sequence in segments according to the time axis. For each pre-defined time sub-interval, the system summarizes all trend potential energy values ​​within that sub-interval to obtain the cumulative trend intensity within that time period. The summarized results of each time period together form a set of trend energy sub-vectors, which have a chronological order in the time dimension and comparative ability in the trend intensity dimension. The system can further perform standardization, weight allocation, or classification mapping operations on this vector set as needed to generate the final trend energy expression vector. This expression vector, as a high-dimensional structured input in the trend identification stage, not only contains a comprehensive representation of trend directionality, trend amplitude, and trend persistence, but also has good model adaptability and classification discrimination ability, providing key data support for achieving high-precision equipment status trend classification and predictive maintenance.

[0075] In practical applications, to ensure that the trend energy expression vector has sufficient temporal resolution and trend discrimination capability, the system divides the trend energy expression sequence into multiple time sub-intervals of uniform length, for example, each sub-interval consists of 10 time sampling points. Taking a trend energy sequence with a total length of 100 points as an example, the system can divide it into 10 consecutive time periods and sum the trend energy values ​​within each time period to obtain the overall trend strength within that time period, thereby generating 10 trend energy sub-vectors. For example, if the trend energy values ​​in the first time period are mostly concentrated at high levels, such as 0.85, 0.88, 0.82, etc., then the trend energy sub-vector of that period will show a high value, indicating that the device status has a significant stable trend during that time period; while if the trend energy values ​​fluctuate greatly and the average is low during a certain time period, then the corresponding trend energy sub-vector will be a low value, reflecting that the trend of that data segment is unclear or that frequent reversals occur. The system can further standardize these sub-vectors, for example, by mapping the overall trend energy range to a fixed interval or introducing weighting coefficients to emphasize the influence of key time periods, ultimately forming a trend energy expression vector in a unified format. This expression vector not only retains the trend persistence and directional features in the original data, but also improves feature compressibility and input consistency through structural integration. It provides directly callable high-dimensional structured input for subsequent trend state classification, offset identification, and prediction model training, greatly enhancing the system's ability to characterize the evolution trend of equipment states and its response speed.

[0076] S4. Decompose the trend energy expression vector into multiple positive and negative axes, transforming the change in each axis into an independent distribution curve, and generate the trend offset index through convolution by superimposing probability increments; as detailed below.

[0077] The trend energy expression vector is deconstructed dimensionally and decomposed axially to construct positive and negative directional subsequences. After obtaining the trend energy expression vector, the system treats it as a structured data vector containing trend intensity expressions across multiple time periods. Each dimension of this vector corresponds to a trend energy value within a fixed time period, possessing a clear temporal sequence attribute. To further identify the bias and evolution direction of the trend within each time period, the system performs axial deconstruction on this vector in this step. Specifically, the system first labels the trend energy value of each dimension with positive and negative directions, determining whether its trend orientation is positive (increase), negative (decrease), or neutral (fluctuation without a clear direction), and accordingly decomposes the overall trend energy expression vector into two independent but structurally corresponding sub-vectors: a positive sub-vector, retaining only the energy values ​​of all trend strengthening segments and setting the rest to zero; and a negative sub-vector, retaining only the energy values ​​of all trend weakening segments and setting the rest to zero. This decomposition method not only preserves the directional attribute of the trend but also forms two trend expression paths with consistent structures and opposing directions. Furthermore, to ensure the stability of subsequent calculations, the system will normalize the positive and negative sub-vectors separately to make them comparable in magnitude, ensuring that the influence of trends in different directions can be analyzed and weighted in parallel.

[0078] Taking a trend energy expression vector of length 8 as an example, assuming the vector is: [0.12, 0.25, -0.08, 0.00, -0.15, 0.30, -0.05, 0.10], it represents the trend energy performance of the device status over 8 consecutive time intervals. The system first determines the positive or negative direction of each value, labeling positive values ​​(such as the 1st, 2nd, 6th, and 8th bits) as "increasing trend", negative values ​​(such as the 3rd, 5th, and 7th bits) as "decreasing trend", and a value of 0 (the 4th bit) as having no obvious trend. Then, the system constructs a positive subvector: [0.12, 0.25, 0.00, 0.00, 0.00, 0.30, 0.00, 0.10], retaining only positive values ​​and setting the rest to 0; at the same time, it constructs a negative subvector: [0.00, 0.00, -0.08, 0.00, -0.15, 0.00, -0.05, 0.00], retaining only negative values ​​and setting the rest to 0. Next, to improve the consistency of subsequent analysis, the system normalizes the two sub-vectors separately, for example, by scaling them according to their maximum absolute values, so that the positive sub-vector is normalized to [0.4, 0.83, 0.00, 0.00, 0.00, 1.00, 0.00, 0.33], and the negative sub-vector is normalized to [0.00, 0.00, 0.53, 0.00, 1.00, 0.00, 0.33, 0.00]. This processing method fully preserves the structure and directionality of the positive and negative trends, and also lays the foundation for subsequent trend probability modeling and trend deviation index construction.

[0079] For both positive and negative sub-vectors, probability density distribution curves are constructed to extract the trend direction change profile. After separating the positive and negative trend energy, the system treats the positive and negative sub-vectors as a set of continuous data sequences with distribution characteristics. Based on their internal data fluctuations and time-period distribution patterns, probability density analysis is performed. In this step, the system uses a sliding window approach to process each sub-vector: the trend energy values ​​within a fixed-length time window are statistically analyzed, the frequency and average amplitude of trend enhancement or weakening within that interval are calculated, and a probability density curve is constructed based on the summarized results of all windows. The positive trend probability density curve describes the probability and intensity of upward evolution of equipment operating status in continuous time, while the negative trend probability density curve reflects the frequency and amplitude concentration range of downward evolution or degradation of operating status. Through this step, the system not only obtains the directional distribution of the trend in the time structure but also constructs a trend description function with probabilistic attributes, transforming the original energy value sequence into a trend behavior distribution. This process realizes the transformation from a time series trend intensity description to a trend behavior space, providing a distribution basis for the next step of trend offset quantification.

[0080] Taking a positive subvector [0.4, 0.83, 0.00, 0.00, 0.00, 1.00, 0.00, 0.33] as an example, the system sets the sliding window length to 3 time points and performs sliding statistics on it. The first window covers positions 1-3, with a trend value of [0.4, 0.83, 0.00], a non-zero frequency of 2 / 3, and an average trend amplitude of (0.4+0.83) / 2 = 0.615; the second window covers positions 2-4, with a value of [0.83, 0.00, 0.00], a non-zero frequency of 1 / 3, and an average amplitude of 0.83; and so on. The system traverses all sliding windows, records the frequency and average amplitude of trend enhancement within each window, and constructs a "probability distribution curve of enhanced trend occurrence" for time position. Similarly, for the negative subvector [0.00, 0.00, 0.53, 0.00, 1.00, 0.00, 0.33, 0.00], the system extracts subsequences such as [0.53, 0.00, 1.00] through a sliding window, calculating a frequency of 2 / 3 and an average amplitude of (0.53+1.00) / 2 = 0.765, gradually generating a "probability density curve of weakening trend". Ultimately, the two probability density curves describe the distribution of trend strengthening and weakening over time, providing input with directional distinction and probabilistic characterization for subsequent trend-shifting convolution. This method effectively captures the temporal structure and amplitude concentration of trend behavior, enabling the system to establish a trend pattern discrimination mechanism based on probabilistic behavior.

[0081] The system performs incremental convolution fusion on the positive and negative trend probability density curves to form a single trend offset exponential function. To unify bidirectional trend behavior into a comparable single index, the system transforms the aforementioned positive and negative probability density curves into trend increment functions and performs cross-directional superposition convolution. Specifically, the system performs time-axis sliding processing on the positive trend probability curve to extract the local trend strengthening degree at each position; then it performs reverse sliding processing on the negative trend curve to extract its local trend weakening strength. Subsequently, at each corresponding time position, the system performs weighted convolution fusion on the strengthening increment of the positive trend and the weakening increment of the negative trend to form a single numerical output, representing the "net trend offset degree" at that time point. When the positive strengthening is much greater than the negative weakening at a certain time point, the offset exponent is positive, and vice versa. The system repeats this operation at all time points, ultimately generating a trend offset exponential sequence. This sequence has a complete time structure, reflecting the combined performance of the trend evolution direction and intensity at each time point during continuous operation of the equipment. Compared to using trend energy values ​​or guidance values ​​alone, the trend deviation index integrates directional attributes, distribution breadth, and rate of change, giving it a stronger ability to identify trends and respond to system changes.

[0082] Taking the previously constructed positive trend probability density curves [0.6, 0.8, 0.2, 0.0, 0.0, 0.9, 0.1, 0.5] and negative trend probability density curves [0.0, 0.1, 0.7, 0.6, 0.9, 0.0, 0.2, 0.3] as examples, the system performs sliding incremental extraction from left to right for the positive curves, selecting a window [0.2, 0.0, 0.0] at position 3, with an average enhancement degree of 0.067; for the negative curves, a window [0.9, 0.0, 0.2] is extracted from right to left, with an average weakening strength of 0.367. The system integrates these two values ​​through a weighted convolution. For example, if the positive weight is set to 0.6 and the negative weight to 0.4, the net trend shift at that location is (0.6 × 0.067) – (0.4 × 0.367) ≈ –0.088, indicating a weaker overall trend and a tendency towards degradation at the current location. The system repeats this operation for all time locations, ultimately generating a trend shift index sequence of length 8, for example: [+0.35, +0.41, –0.09, –0.20, –0.28, +0.47, –0.02, +0.19]. This sequence indicates the strength and direction of the trend change of the equipment at each time point; positive values ​​indicate a strengthening of the operating state, while negative values ​​indicate a tendency towards degradation or performance decline. Through this trend shift index, the system not only achieves a fusion expression of bidirectional trend behavior but also possesses the ability to dynamically track and respond to the trend evolution process, significantly improving the timeliness and targeting of predictive maintenance.

[0083] The system enhances the stability of the trend deviation index sequence and outputs it as a feature vector, serving as a key indicator for subsequent evaluation. After generating the trend deviation index sequence, the system performs moving average and outlier smoothing to remove spikes or sharp reversals caused by local instantaneous fluctuations, enhancing the stability of trend identification. Subsequently, the system divides the deviation index sequence into several logical segments based on a set time period and summarizes and extracts features from the deviation index values ​​within each segment, including multiple descriptive indicators such as maximum value, average value, positive-to-negative fluctuation ratio, and slope of change. Finally, these local deviation features are combined into a trend deviation feature vector with a unified dimension and model input format for use in subsequent state classification, health level assessment, or early warning strategy formulation. In addition, the system can construct a risk accumulation indicator based on the overall dynamic change rate of the trend deviation index to identify "trend reversal precursors" or "long-term deviation deposition sections," thereby further enhancing the foresight and accuracy of predictive maintenance capabilities. The generation of this trend deviation vector marks the completion of the entire closed loop from the original state change path to the structured trend expression, enabling the entire system to have state recognition and decision-making capabilities oriented towards trend evolution.

[0084] S5. Establish the mapping interval for the trend deviation index, construct an asymmetric distribution model using the critical change range in historical samples, set a three-segment state classification interval, and map and classify the current trend deviation index; details are as follows:

[0085] The system collects historical equipment operation data and extracts corresponding trend offset indices to establish a complete sample base to support the construction of trend classification mapping rules. In practical applications, the system first collects historical equipment data covering multiple operating states. This data must originate from long-term operation monitoring records in real industrial scenarios and encompass the complete state evolution process from stable equipment performance to mild degradation and then to severe failure. The system processes this historical data line by line using the previously constructed trend energy expression method, generating a trend offset index value corresponding to each time point. Simultaneously, the system retrieves manual maintenance records, alarm records, or quality control results corresponding to these historical trend offset indices and aligns them with timestamps to identify whether human intervention or system anomaly indicators exist when the trend offset index changes. In this way, the system establishes a correlation mapping between "trend offset index changes" and "actual state transition events," providing a real basis for the subsequent formulation of classification standards. Furthermore, the system labels operating events, such as "remaining stable," "mild degradation but no failure," and "failure triggered," and summarizes and analyzes the trend offset index intervals of these labels to extract typical offset ranges corresponding to critical changes. The resulting dataset includes both the offset values ​​themselves and their operational meaning in the actual operation and maintenance environment, laying a solid foundation for the construction of the distributed model.

[0086] Based on the correspondence between trend offset indices and actual state labels, an asymmetric distribution model is constructed, and a three-segment classification interval is defined for state identification. After forming a historical trend offset index sample covering multiple state labels, the system statistically consolidates these indices and finds that the distribution of the trend offset indices is not symmetrical. For example, in most scenarios, a slight negative offset can trigger a warning of equipment performance degradation, while a large positive offset does not necessarily indicate risk. Therefore, the system no longer uses the average symmetric segmentation method, but instead constructs an asymmetric trend offset distribution model. The system classifies all offset index samples according to their state and statistically analyzes their distribution boundaries, peak frequency bands, and critical concentration regions under different operating states. For example, the system may find that the offset indices of slightly degraded states are mostly concentrated between -0.10 and -0.25, while the offset values ​​of severely degraded states are concentrated in the range below -0.30; meanwhile, the offset values ​​of performance-enhanced states are mainly distributed between +0.15 and +0.40. Based on this finding, the system divides the system into three non-overlapping intervals: the first is the central interval, representing stable operation, mainly covering the range from -0.10 to +0.15; the second is the negative degradation interval, extending from -0.10 to lower, representing different levels of performance degradation and failure risk; the third is the positive enhancement interval, extending from +0.15 upwards, reflecting potential performance improvement or high-load operation trends. This three-segment structure ensures that each trend has a clear direction, while introducing asymmetric boundary control to control the probability of misjudgment, making the system practically adaptable.

[0087] The system maps the currently collected trend offset index to preset classification intervals in real time, completing the linkage between state level judgment and predictive maintenance response. After completing the interval modeling of the trend offset index, the system enters the real-time mapping stage. Whenever the equipment generates new state data during operation, a specific value of trend offset index is obtained after data preprocessing, trend identification, normalization preservation, trend potential energy conversion, and trend offset generation. The system first reads the current offset index value and compares it with the preset three-segment classification interval: if the value falls into the central stable interval, it is determined that the current equipment operation status is normal and no maintenance intervention is required; only the standard monitoring frequency is maintained. If the value falls into the negative degradation interval, the system immediately enters the warning state and further judges the depth position of the offset value in the degradation interval, and decides whether to push maintenance suggestions based on the duration and historical comparison trends. If the value falls into the positive enhancement interval, it means that the equipment may be in a high load or structural stress increase stage. The system will judge whether its trend has risk attributes and adjust the normalization scale or time window as necessary to improve the model sensitivity. Through this classification and mapping mechanism, the trend deviation index not only has real-time discrimination capabilities but also dynamic control capabilities. Its numerical changes directly guide the predictive maintenance system to respond, effectively enabling early identification, early prediction, and early intervention of potential equipment risks.

[0088] S6. Based on the continuous rate of change of the trend deviation index, the local structure of the scalar of the trend change is deduced, and the time level weights and normalization function parameters are dynamically updated through the difference inversion mechanism to realize the online correction of the trend determination parameters.

[0089] Based on the continuous changes in the trend offset index, the system identifies abrupt shifts in offset rate within key time windows and initializes an offset tracing mechanism. In predictive maintenance scenarios for actual equipment, the trend offset index should exhibit a relatively continuous, slowly changing structure over time. However, in some cases, when a potential weak abnormal evolution trend emerges within the equipment, the trend offset index may show accelerated changes with continuous increases or decreases over a certain period. To capture these changes, the system first constructs a continuous tracking mechanism, performing time-sliding window processing on the trend offset index sequence and recording the direction and magnitude of change for each time period. By calculating the directional consistency and trend change intensity between multiple adjacent time segments, the system identifies time periods in the trend offset sequence where significant accelerated evolution occurs. These time periods are defined as "trend anomaly windows," representing trend information in the state data that is not fully expressed by the current modeling structure. After this identification, the system uses these "trend anomaly windows" as key sections for subsequent reverse structure tracing and initiates the offset tracing mechanism. It performs time-by-time comparative analysis on all normalized input data within the anomaly window to deduce whether the current trend offset change originates from the accumulation of errors or attenuation of the directional change scalar. This step lays the foundation for the target window for subsequent difference inversion and online correction, ensuring that the judgment and adjustment process has clear boundaries and positioning basis.

[0090] A local structural comparison relationship of the scalar guidance change is established, and a difference inversion mechanism is executed to locate the root cause of trend deviation. After locating the abnormal trend deviation window, the system performs point-by-point structural analysis on the scalar guidance change structure corresponding to that window. The goal of this implementation step is to identify the missing directional expression that may have been ignored or compressed in the pre-normalization stage behind the trend deviation. The system first retrieves the original pre-normalization state parameter sequence that is completely consistent with the time period of the abnormal window, and reconstructs the difference sequence and direction label sequence of that time period to obtain the ideal scalar guidance change structure that should be formed. Subsequently, the current scalar guidance change sequence is compared one by one with the reconstructed guidance structure to analyze the deviation between the two in terms of slope direction, direction consistency factor, and continuity expression. If it is found that the existing guidance structure has frequent fluctuations in the trend direction, insufficient guidance continuity, or a change amplitude that lags significantly behind the change of state parameters, it can be determined that the expression of the scalar guidance change is distorted, leading to an increase in the error of subsequent trend deviation judgment. The system further quantifies the aforementioned structural differences into a set of trend expression shift difference signals, and aggregates them over time to construct a trend shift difference map. The map clearly marks the points where each directional expression shift is inconsistent with the trend change. This map serves as a direct reference for the system's dynamic correction, guiding the precision adjustment of the normalization function and time weight allocation parameters, and is a key support point for establishing a trend determination feedback loop.

[0091] Based on the difference map results, the system implements normalization function parameter correction and time hierarchy structure adjustment to achieve online evolution of trend judgment capability. After obtaining the difference map of trend expression shift, the system enters the dynamic adjustment stage of trend judgment parameters. First, regarding the adjustment of the normalization function boundary, the system performs aggregate analysis on the time periods in the difference map where misjudgments frequently occur, and statistically analyzes the boundary distance distribution of these points in the original normalization process. If it is found that most of these missing trend expression points are concentrated in the upper and lower boundary regions of the normalization function, it indicates that the original boundary setting is too tight and cannot accommodate weak but continuous trend changes. At this time, the system automatically expands the normalization function boundary in this time period appropriately to improve the expression fidelity of trend change amplitude in normalization processing. At the same time, the system performs a redistribution operation on the sampling window weight allocation strategy in the time hierarchy structure, transferring more sampling weights to the time period corresponding to the trend shift anomaly window, and reducing the weights of other slowly changing time periods to improve the overall modeling responsiveness to trend-sensitive areas. Furthermore, the system establishes a parameter evolution caching mechanism to record the boundary expansion magnitude and weight adjustment ratio during each adjustment process, and introduces a trend inertia factor to smooth the weight update process, preventing over-adjustment due to noise or short-term fluctuations. The entire parameter correction process is triggered immediately after the system detects that the rate of trend deviation exceeds the threshold, possessing non-interrupted online update capabilities. It can continuously adapt to the trend change structure of equipment operating status, significantly enhancing the stability and predictive sensitivity of the predictive maintenance system in long-term deployment environments.

[0092] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, as a computer program product. A computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, all or part of the processes or functions according to the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired or wireless means (e.g., infrared, wireless, microwave, etc.). A computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that includes one or more sets of available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media. Semiconductor media can be solid-state drives.

[0093] It should be understood that in the various embodiments of this application, the order of the above-mentioned processes does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of this application.

[0094] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0095] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interfaces, devices, or units, and may be electrical, mechanical, or other forms.

[0096] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0097] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0098] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A predictive maintenance method for equipment with anomaly prediction capabilities, characterized in that, Specifically, the following steps are included: S1. Construct a multi-time-level data acquisition window to generate multiple non-overlapping time intervals for the device status parameters. The device status parameters are real-time status parameter data collected during device operation, used to characterize the changes in the device's operating status. Perform differential calculations within each time interval and generate a directional change scalar based on the direction of time series change. S2. Construct a directional continuity factor based on the directional change scalar and historical statistical fluctuation values. The historical statistical fluctuation values ​​are obtained by statistically calculating the state parameters collected by the equipment under fault-free operation. They are used to characterize the fluctuation range of the state parameters under normal operating conditions. The directional continuity factor is used to control the numerical recalibration function in the normalization process to avoid compressing the continuous change trend in the normalization process. S3. The normalized data is transformed into a continuous curvature space. The data is a sequence of state parameters arranged in time order. By analyzing the local slope continuity of the data change path, a trend potential energy is formed, and a trend energy expression vector is generated by performing path integration based on the trend potential energy. S4. Decompose the trend energy expression vector into multiple positive and negative axes, and transform each axial change into an independent distribution curve. The independent distribution curves correspond to the trend strengthening direction and the trend weakening direction, respectively. Generate the trend offset index by convolution with probability increments. S5. Establish the mapping interval of the trend deviation index, construct the correspondence between the trend deviation index and the equipment operating status based on the historical operating data of the equipment, and use it to characterize the changing trend of the equipment operating status. Use the critical change range in the historical samples to construct an asymmetric distribution model, set a three-segment state classification interval, and map and classify the current trend deviation index. S6. Based on the continuous rate of change of the trend deviation index, the local structure of the scalar of the trend change is deduced. Based on the time series of the trend deviation index, abnormal change segments are identified. The time level weights and normalization function parameters are dynamically updated through the difference inversion mechanism to realize the online correction of the trend judgment parameters.

2. The predictive maintenance method for equipment with anomaly prediction capability according to claim 1, characterized in that, S1 specifically includes: A basic sampling period is set, and multiple sets of time windows of different lengths and non-overlapping time windows are constructed in the time dimension. Each time window is divided into multiple sampling points according to the basic sampling period, forming a data hierarchical structure with short-term, secondary-term and medium-to-long-term scales. Within each time window, the difference between adjacent data is calculated point by point for the corresponding state parameter data sequence to generate a complete difference sequence, which is used to characterize the direction and magnitude of change within that time period. Based on the statistical proportion of positive and negative directions of the difference sequence, and combined with the absolute average magnitude of the difference, a directional change scalar is generated to represent the directional consistency and trend strength of state changes within the time window.

3. The predictive maintenance method for equipment with anomaly prediction capability according to claim 1, characterized in that, S2 specifically includes: Historical status parameter data of the equipment under fault-free conditions are collected, and multiple sliding time windows are constructed based on a fixed sampling period. The maximum value, minimum value, mean and standard deviation within each sliding time window are calculated to form a set of historical statistical fluctuation values ​​for reference. The ratio of the directional change scalar in the current time period to the corresponding historical standard deviation is calculated to obtain the directional continuity factor used to represent the trend direction and the relative change magnitude. The maximum or minimum boundary of the normalization function is adjusted according to the positive or negative value of the direction continuity factor so that the normalized state parameter data retains the original change direction characteristics. The system performs a consistency comparison of the change direction before and after normalization on the normalized state parameter data. When the consistency ratio is lower than the set threshold, it returns to adjust the continuity factor of the direction and re-normalizes.

4. The predictive maintenance method for equipment with anomaly prediction capability according to claim 1, characterized in that, S3 specifically includes: By treating the normalized state parameter sequence as a continuous change path, we analyze the change direction and bending intensity between each data segment in time sequence, and construct a curvature space structure with time traceability. Based on the constructed curvature structure, the direction of data change is determined to be continuous in each continuous time segment to form a trend continuity label, and a weight corresponding to the trend direction is assigned to each trend segment. Based on the curvature characteristics and trend continuity weights at each time point, a trend potential energy sequence is generated. The trend potential energy value is used to represent the trend direction and change magnitude of that point in the state evolution process. The trend potential energy sequence is accumulated and integrated according to a preset time period to form a trend energy expression vector containing multiple dimensions, which is used for subsequent state trend analysis and classification identification.

5. The predictive maintenance method for equipment with anomaly prediction capability according to claim 1, characterized in that, S4 specifically includes: The trend energy expression vector is decomposed into positive sub-vectors and negative sub-vectors according to the numerical signs of each dimension, which correspond to the expression paths of the strengthening trend and weakening trend of the state parameters in each time interval, respectively. Based on the positive and negative sub-vectors, trend probability density distribution curves are constructed to reflect the distribution of the probability of occurrence and the magnitude of trends in each direction over time. The probability density curves of positive and negative trends are converted into trend increment functions, and weighted convolution is performed at the corresponding time positions to form a trend offset index, which is used to characterize the direction and intensity of net trend change.

6. The predictive maintenance method for equipment with anomaly prediction capability according to claim 1, characterized in that, S5 specifically includes: Collect historical operating data of the equipment, generate a trend deviation index for each historical time point, compare and label it with the status label corresponding to that time point, and build a correspondence between the trend deviation index and the equipment operating status. Based on the aforementioned correspondence, the distribution characteristics of the trend deviation index under various states are statistically analyzed, an asymmetric distribution model is established, and the trend deviation index is divided into three state classification intervals: central stable interval, negative degradation interval, and positive enhancement interval. The currently collected trend deviation index is mapped to the classification interval in real time. The equipment operating status is determined based on the mapping result, and predictive maintenance decisions are executed in conjunction with these decisions, including risk warnings, maintenance suggestion pushes, or monitoring frequency adjustment operations.

7. The predictive maintenance method for equipment with anomaly prediction capability according to claim 1, characterized in that, S6 specifically includes: Based on the continuous rate of change of the trend deviation index, anomaly windows of trend deviation are identified, and a deviation tracing mechanism is initialized to locate the time period in which potential trend expression is missing. The scalar structure of the guidance change within the trend offset anomaly window is compared with the time-period difference, the ideal guidance structure for the corresponding time period is reconstructed, and a trend expression offset difference map is generated based on the structural deviation. Based on the trend expression offset difference map, the weight allocation of the normalization function boundary parameters and the time level acquisition window is dynamically adjusted, and the parameter evolution process of each round of adjustment is recorded to achieve non-interrupted online correction of the trend determination parameters.

Images