Digital-twin-based device fault prediction and health assessment method and system

Abstract

The application provides a device fault prediction and health degree evaluation method and system based on digital twinning. The method comprises the following steps: generating a digital twinning reference result strictly corresponding to a current operation segment by combining a digital twinning model with device state data and working condition identification data; secondly, generating a fault prediction result based on the deviation of the actual state and the reference result and the evolution relationship of the adjacent time slices; subsequently, constructing an evaluation basis result by combining the prediction result change relationship on the continuous time slices to distinguish between continuous degradation and instantaneous fluctuation; finally, mapping the evaluation basis result to a health degree base value and dynamically correcting the trend to output a health degree dynamic evaluation result. The application effectively solves the problem that the device state fluctuation and the real degradation are difficult to separate under complex working conditions by constructing a continuous processing chain from the digital twinning reference to the health degree evaluation, thereby improving the sensitivity of fault prediction and the stability of health degree evaluation.

Description

Technical Field

[0001] This invention belongs to the field of industrial control technology, and in particular relates to a method and system for equipment fault prediction and health assessment based on digital twins. Background Technology

[0002] As industrial equipment evolves towards higher speeds, continuous operation, and complex operating conditions, its operational status is increasingly influenced by a combination of factors, including load variations, speed switching, control parameter adjustments, process cycle variations, and natural component aging. This results in the same equipment exhibiting significantly different, yet not necessarily faulty, operating characteristics at different points in time. Against this backdrop, fault prediction and health assessment for industrial equipment have evolved from traditional methods relying on fixed thresholds, periodic maintenance, and experience-based judgment to a comprehensive approach combining online monitoring, condition analysis, and virtual-physical mapping modeling. In particular, the introduction of digital twin technology allows for the creation of a virtual operational reference corresponding to the actual equipment, providing a new foundation for understanding its condition under complex operating conditions.

[0003] In existing technologies, one type of approach mainly extracts features directly from monitoring data such as vibration, current, temperature, and pressure, and then uses rules, statistical models, or learning models to identify anomalies. Although this approach is relatively straightforward, the original monitoring signals are highly sensitive to changes in operating conditions, and often misjudge the state fluctuations caused by the transition from normal operating conditions as fault symptoms, making it difficult to maintain stable judgment under multiple operating conditions. Another approach, while introducing digital twin models to compare actual and virtual states to improve the interpretability of fault identification, still faces two prominent problems in engineering applications. First, many approaches utilize digital twin results only at the level of single-shot deviation comparisons, lacking the ability to organize condition-related references, fault prediction, and health assessment into a continuous computational chain. This renders the digital twin merely an auxiliary comparison, not truly an integral part of the diagnostic process. Second, many health assessment methods directly map abnormal or predicted quantities at a specific moment to scoring results, failing to fully utilize the evolutionary characteristics of prediction results over continuous time slices. This allows short-term disturbances, local spikes, and sampling fluctuations to easily propagate directly to the health results, leading to unstable assessments, unclear trends, and insensitivity to early, slow degradation. Especially in industrial settings, early degradation often does not manifest as drastic anomalies but rather as a continuous deviation from the current condition reference state and a slow accumulation over continuous time slices. Without a continuous processing mechanism designed around this characteristic, it is difficult to balance the sensitivity of fault prediction with the stability of health assessment. Summary of the Invention

[0004] This invention discloses a method and system for predicting equipment failure and assessing health based on digital twins, in order to solve the problems mentioned in the background art.

[0005] To achieve the above objectives, the first aspect of the present invention provides a method for equipment fault prediction and health assessment based on digital twins, the method comprising: Acquire equipment status data and operating condition identification data of industrial equipment, input the equipment status data and operating condition identification data into a pre-deployed digital twin model, and generate a digital twin reference result that strictly corresponds to the current operating segment; The actual state vector, which is synchronously saved with the digital twin reference result, is retrieved in the edge computing unit. The operating condition normalization deviation vector is calculated based on the actual state vector and the digital twin reference result. The fault prediction result is generated by combining the adjacent time slice evolution relationship of the operating condition normalization deviation vector. Based on the fault prediction results on continuous time slices, an evaluation basis is constructed that includes the current prediction level, trend normalization increment, and second-order difference curvature constraint. The evaluation criteria results are used to generate a health baseline value through exponential decay mapping. The health baseline value is then dynamically corrected by combining the changing trends of the evaluation criteria results in adjacent time slices, and the dynamic health evaluation result is output.

[0006] Furthermore, in the step of inputting the equipment status data and operating condition identification data into a pre-deployed digital twin model to generate a digital twin reference result that strictly corresponds to the current operating segment, the digital twin model includes an operating condition mapping layer and a status response layer; the operating condition mapping layer is used to generate current operating condition parameters based on the operating condition identification data, and the status response layer is used to generate a digital twin reference result based on the current operating condition parameters and the equipment status data.

[0007] Further, the step of calculating the normalized deviation vector of the operating condition based on the actual state vector and the digital twin reference result specifically involves: calculating the difference between the actual state vector and the digital twin reference result, and dividing the difference by the denominator, which includes the sum of the absolute values ​​of the preset stability term and the digital twin reference result, to obtain the normalized deviation vector of the operating condition.

[0008] Furthermore, the step of generating fault prediction results by combining the evolution relationship of adjacent time slices of the normalized deviation vector of the operating condition specifically includes: weighted summation of the normalized deviation vector of the current time slice and the fault-sensitive weight row vector; calculating the trend difference between the normalized deviation vector of the current time slice and the previous time slice and combining it with the trend weight row vector; and outputting the above two calculation results as a single-value fault prediction result through the logic mapping layer.

[0009] Furthermore, the step of constructing the evaluation basis result including the current prediction level, trend normalization increment, and second-order difference curvature constraint specifically includes: calculating the trend normalization increment of the fault prediction result of the current time slice and the fault prediction result of the previous time slice; calculating the absolute value of the second-order difference of the fault prediction results of the current time slice, the previous time slice, and the time slice before that; weighting and combining the fault prediction result of the current time slice, the trend normalization increment, and the absolute value of the second-order difference, and performing non-negative truncation to generate the evaluation basis result.

[0010] Furthermore, the step of generating a health baseline value by exponential decay mapping of the evaluation basis result specifically involves multiplying the evaluation basis result by a health scale adjustment coefficient and taking a negative exponent to generate the health baseline value for the current time slice.

[0011] Furthermore, the step of dynamically correcting the health baseline value by combining the changing trend of the evaluation results of adjacent time slices specifically involves: calculating the difference between the evaluation results of the current time slice and the evaluation results of the previous time slice; if the difference is greater than zero, then using the product of the difference and the trend adjustment coefficient as the denominator correction term to perform bounded correction on the health baseline value, thereby generating the final dynamic health evaluation result.

[0012] Furthermore, the equipment status data includes at least one of vibration characteristics, current characteristics, temperature characteristics, and pressure characteristics; the operating condition identification data includes at least one of speed range number, load range number, process section number, or recipe number.

[0013] Furthermore, the step of constructing an evaluation basis result that includes the current prediction level, trend normalization increment, and second-order difference curvature constraint based on the fault prediction results on continuous time slices is specifically used to distinguish between continuous degradation trends and instantaneous fluctuation disturbances, so as to weaken the impact of local spikes and short-term disturbances on the evaluation results.

[0014] A second aspect of the present invention provides a digital twin-based equipment fault prediction and health assessment system, the system comprising: The data acquisition module is used to acquire equipment status data and operating condition identification data of industrial equipment, input the equipment status data and operating condition identification data into a pre-deployed digital twin model, and generate a digital twin reference result that strictly corresponds to the current operating segment. The fault prediction module is used to retrieve the actual state vector that is synchronously saved with the digital twin reference result, calculate the normalized deviation vector of the operating condition based on the actual state vector and the digital twin reference result, and generate the fault prediction result by combining the evolution relationship of adjacent time slices of the normalized deviation vector of the operating condition. The evaluation basis module is used to construct an evaluation basis result that includes the current prediction level, trend normalization increment, and second-order difference curvature constraint based on the fault prediction results on continuous time slices. The dynamic evaluation module is used to generate a health baseline value by using the evaluation basis results through exponential decay mapping, and dynamically correct the health baseline value by combining the changing trend of the evaluation basis results of adjacent time slices, and output the dynamic evaluation result of health.

[0015] The beneficial technical effects of the present invention are at least as follows: To address the aforementioned issues, this invention provides a method and system for equipment fault prediction and health assessment based on digital twins. First, a reference result strictly corresponding to the current operating segment is generated using a digital twin model combined with equipment status data and operating condition identification data. This ensures that subsequent analysis focuses on the equipment's deviation from the current operating condition reference state, rather than the original state amplitude detached from the semantics of the operating condition, thus providing a unified reference for fault prediction under complex operating conditions. Building upon this, the invention further generates fault prediction results based on the deviation between the actual state and the reference state and its evolution relationship over adjacent time slices. This ensures that the fault prediction reflects both the current degree of deviation and whether the deviation is continuing to develop, thereby improving the ability to identify early, slow degradation. Subsequently, instead of directly mapping the fault prediction results to health, the invention continues to construct assessment basis results by combining the changes in prediction results over continuous time slices. This allows the assessment process to distinguish between continuous degradation and instantaneous fluctuations, reducing the impact of local spikes and short-term disturbances on subsequent assessments. Finally, the invention converts the assessment basis results into dynamic health assessment results, ensuring that health not only reflects the current equipment state but also the impact of degradation trends on future health. Through this design, the present invention connects digital twin reference, fault prediction, evaluation basis and dynamic health assessment into a coherent and progressively convergent overall solution, enabling digital twin to truly participate in the core calculation process of fault prediction and health assessment. This not only improves the pertinence of fault prediction under complex working conditions, but also enhances the stability, continuity and engineering usability of health results, making it more suitable for online monitoring, predictive maintenance and operation decision-making scenarios of industrial equipment. Attached Figure Description

[0016] The present invention will be further described with reference to the accompanying drawings, but the embodiments in the drawings do not constitute any limitation on the present invention. For those skilled in the art, other drawings can be obtained based on the following drawings without creative effort.

[0017] Figure 1 This is a flowchart of the device fault prediction and health assessment method based on digital twins according to the present invention.

[0018] Figure 2This is a framework diagram of the equipment fault prediction and health assessment system based on digital twins of the present invention. Detailed Implementation

[0019] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0020] In one or more embodiments, such as Figure 1 As shown, a method for equipment failure prediction and health assessment based on digital twins is disclosed, the method comprising the following: S1: Acquire equipment status data and operating condition identification data of industrial equipment, input the equipment status data and operating condition identification data into a pre-deployed digital twin model, and generate a digital twin reference result that strictly corresponds to the current operating segment.

[0021] Specifically, the task of step one is to generate a digital twin reference result. Industrial equipment exhibits different state characteristics under different operating stages, load levels, and control commands. Therefore, the same set of equipment state data only has interpretable meaning when placed within the corresponding operating condition. The processing method used here is to feed the equipment state data collected on-site and the operating condition identification data into a pre-deployed digital twin model, so that the digital twin model outputs a reference state result under the current operating condition. The subsequent steps do not use the original state data itself, but rather the reference result that corresponds one-to-one with the current operating condition. Therefore, this step is actually establishing a standard reference under the current operating condition for subsequent fault prediction.

[0022] Inputs include equipment status data and operating condition identification data. Equipment status data comes from existing acquisition links in the industrial equipment, typically generated by sensors or control drive modules installed on the equipment itself, and then read by controllers, acquisition terminals, or edge computing units. For example, for rotating equipment, equipment status data can come from vibration acquisition channels on bearing housings, current sampling channels in drive cabinets, or status quantities output by the spindle encoder; for pump equipment, equipment status data can come from pump body vibration acquisition channels and drive motor status sampling channels; for CNC spindles, equipment status data can come from spindle current channels and spindle vibration channels. Operating condition identification data comes from the equipment control system and is used to identify which type of operating condition the current operating segment belongs to. Specifically, it can be given by PLC registers, host computer process segment number, recipe number, speed range number, or load range number. The key here is not to introduce many data categories, but to ensure that each set of equipment status data corresponds to a specific operating condition, so that the reference results output by the digital twin model strictly correspond to the on-site conditions.

[0023] During on-site execution, a data segment aligned with the control cycle is first established within the edge computing unit. This data segment consists of equipment status data and operating condition identification data within the same time window. The equipment status data is concatenated into a status vector according to the acquisition order, while the operating condition identification data is mapped to an operating condition parameter vector based on the equipment control logic. To ensure consistent numerical scaling of data from different sources during computation, the status vector and operating condition parameter vector undergo uniform scaling before entering the digital twin model. The processed results are then combined into the input vector. The digital twin model is pre-deployed in edge computing units or industrial servers. The model itself consists of two consecutive parts: the first part is the operating condition mapping layer, which converts operating condition identification data into current operating condition parameters; the second part is the state response layer, which generates reference results based on the current operating condition parameters and equipment status data. The operating condition mapping layer can be pre-configured according to equipment type, for example, configuring parameter groups for fans according to speed range and load range, configuring parameter groups for CNC spindles according to machining stage, and configuring parameter groups for pumps according to flow range. The state response layer adopts a three-layer fully connected network. The first layer expands the input vector to a unified feature space, the second layer extracts the effective state associations under the operating condition, and the third layer outputs the digital twin reference result. The calculation relationship is as follows: ; in, The input vector is obtained by combining equipment status data and operating condition parameter vectors, and is specifically spliced ​​by the edge computing unit within the same control cycle. The equipment status data is read by field sensors and drive sampling links, and the operating condition parameter vector is obtained by looking up the table through the operating condition mapping layer from the operating condition identification data. , , This represents the parameter matrix of a three-layer fully connected network. These parameters are obtained by training with historical samples during the normal operation phase of the device and deployed to the field, and are directly invoked during the online execution phase. The activation layer is represented by an element-wise nonlinear mapping, which is used to characterize the nonlinear relationship between equipment state and operating condition. This represents the digital twin reference result, used to characterize the reference state that the equipment should exhibit under the current operating conditions. In actual deployment, this calculation process is executed continuously online. Taking a centrifugal pump as an example, the edge computing unit reads the pump body vibration characteristics and drive state characteristics within a certain control cycle, and simultaneously reads the current flow range number from the PLC. The operating condition mapping layer first selects the corresponding parameters based on the flow range number, and then combines them with the vibration characteristics and drive state characteristics to form an input vector. The state response layer then outputs a reference result for that flow range. Taking a CNC spindle as another example, the system reads the spindle current and vibration characteristics, and simultaneously reads the current machining stage number. For example, different operating parameters are used for the roughing and finishing stages, ultimately obtaining the corresponding reference results for each stage. This is how it is generated. It is always tied to the current operating condition. Subsequent steps deal with the relationship between the actual state and the current operating condition reference state, rather than a simple state value detached from the operating condition.

[0024] This step outputs the digital twin reference result. . The reference state vector corresponding to the current input sample is calculated by the digital twin model under the current operating conditions. Subsequent steps will directly use this result for fault prediction. By introducing equipment status data and operating condition identification data into the digital twin model, this step forms a reference result that corresponds one-to-one with the current operating segment. This result unifies the expression of the equipment's operating condition and current state within the same computational framework, enabling subsequent fault prediction to focus on the deviation from the current operating condition rather than the original state amplitude. This is more suitable for the actual operating characteristics of industrial equipment under multiple operating conditions.

[0025] S2: Retrieve the actual state vector that is synchronously saved with the digital twin reference result in the edge computing unit, calculate the normalized deviation vector of the operating condition based on the actual state vector and the digital twin reference result, and generate the fault prediction result by combining the evolution relationship of adjacent time slices of the normalized deviation vector of the operating condition.

[0026] Specifically, step two directly inherits the digital twin reference results output from step one. The study focuses on whether the discrepancy between the expected and actual state of the equipment under current operating conditions has already led to signs of degradation. Step one maps equipment state segments and operating condition identification segments into a digital twin model and outputs reference results that are strictly paired with that time slice. This step uses the same time slice index to retrieve the data from the cache of the edge computing unit. Synchronously saved actual state vector This forms a set of strictly homologous, identical, and simultaneous comparative objects. It's not a newly introduced type of field data, but rather data generated in step one. The corresponding records of the state segments that have already been calculated are stored in the cache, such as the vibration characteristics and drive current characteristics of the fan bearing housing, the vibration characteristics and drive state characteristics of the centrifugal pump body, and the current and vibration characteristics of the CNC spindle. Because The current operating conditions have already been incorporated into the digital twin computing framework. This step will not repeat the coding of the operating conditions, but will directly utilize... By constraining the differences in field conditions to the semantics of the current operating condition, the resulting fault prediction results are obtained. Only then can it be explained for various operating scenarios of industrial equipment.

[0027] This step first constructs the normalized deviation vector for the operating condition. The original source of this formula is the mathematical expression for relative error. Its classic form uses a reference value as the denominator to construct a deviation ratio, used to eliminate scale inconsistencies when directly comparing objects of different magnitudes. Applying this approach to the digital twin scenario of industrial equipment, the reference value is naturally chosen as the digital twin reference result output in step one. ,because This corresponds to "the state the equipment should exhibit under the current operating conditions"; further considering that the equipment state components may have components close to zero, a preset stability term is added to the denominator. and to Taking the absolute value ensures that the normalized deviation remains stable under different operating conditions and state amplitudes. The resulting operating condition normalized deviation vector is: ; in, This represents the actual state vector of the device corresponding to the current time slice, determined by the edge computing unit based on... The time slice index is read from the state cache used in step one; This indicates that the digital twin reference result output in step one is obtained by calculating the expected state of the equipment under the current operating conditions using the digital twin model; This indicates a pre-defined stable term, which is written as a fixed parameter into the edge computing unit during system deployment; This represents a vector obtained by taking the absolute value of each element of the reference result, used to provide a scale reference consistent with the current operating conditions for each state component; This represents the normalized deviation vector of the operating conditions, used to characterize the proportion of deviation between the actual state of the equipment and the current reference state of the operating conditions. The changes in this formula after its derivation from the classical relative error form to the industrial digital twin scenario are mainly reflected in three aspects: First, the fixed reference quantity in the traditional relative error is replaced with the dynamically generated operating condition-related reference result from step one. Secondly, introduce into the denominator The reference component, which is close to zero, remains stable; thirdly, the denominator adopts... Instead of directly adopting This approach makes the deviation calculation independent of the direction of the reference state and only related to the reference amplitude. After this processing, the deviation expression no longer merely reflects the numerical difference, but rather the "proportion of deviation relative to the expected state under the current operating condition," which is particularly crucial for fault prediction when equipment operates across different operating conditions. Taking a centrifugal pump as an example, the normal amplitude of the pump body vibration characteristics is inherently different in the high-flow-rate and low-flow-rate ranges. If the absolute difference is directly compared, the same difference in the two ranges will be given the same interpretation; however, by using... Then, this difference will be automatically normalized according to the reference state amplitude under the current operating conditions, so that the deviations in the high flow range and the low flow range are comparable.

[0028] In obtaining This step then generates fault prediction results. The original source of this formula comprises two parts: one part comes from the linear discriminant method, which uses weighted summation to extract the component most sensitive to the target; the other part comes from the first-order backward difference method in numerical analysis, which uses the difference between the current quantity and the quantity at the previous time step to describe the evolution direction. Combining these two classic approaches, the linear discriminant part is used to highlight the deviation component most relevant to equipment degradation, while the backward difference part is used to reinforce the continuous growth trend of the deviation between adjacent time slices, thus allowing early degradation to be included in the prediction results when the amplitude is still small but the trend has already formed. To enable steps three and four to be calculated directly using time series scalars, the fault prediction results here... Defined as the single-value prediction value corresponding to the current time slice. The specific calculation relationship is as follows: ; in, This represents a fault-sensitive weight row vector with dimension . The results are derived from offline training before system deployment. During offline training, normalized deviation samples of operating conditions between the normal operation stage and the confirmed early degradation stage of the equipment are collected according to equipment type. Using these samples as input and the corresponding degradation labels as targets, the contribution weight of each deviation component to degradation prediction is obtained through constrained linear regression or logistic regression, and the trained weight row vector is solidified into the edge computing unit. This represents the normalized deviation vector of the operating conditions in the current time slice, with dimension . The result is obtained from the previous formula; This represents the normalized deviation vector of the operating conditions in the previous time slice, also with dimension 1. The data is stored in the sliding buffer by the edge computing unit in chronological order. This represents the trend weight row vector, with dimension 1. It is used to control the contribution of the changing trends of each deviation component to the prediction, and is configured according to the equipment type during deployment; This represents the logic mapping layer, which specifically uses logic mapping to ensure that the output falls within a unified range. This represents the fault prediction result, a single-value prediction value corresponding to the current time slice, used to characterize the predicted intensity of equipment degradation symptoms under the current operating condition. The changes in this formula derived from the classical linear discriminant form are reflected in two aspects: first, the discriminant object is no longer the original state vector, but the operating condition normalization deviation vector jointly constructed in step one and this step. Second, in the linear discriminant term Add on the basis This allows the prediction results to simultaneously reflect both the "magnitude of the current deviation" and the "direction of deviation evolution over time." For industrial equipment, many early failures do not initially manifest as large anomalies, but rather as a slow increase in deviation over multiple consecutive time slices. This term can extract this "continuously growing but individually small" sign of degradation.

[0029] Depending on the specific execution method at the application level, the edge computing unit obtains [the necessary information] in the current time slice. Then, first read the corresponding state vector from the cache. Then calculate according to the above formula. Then read the data saved from the previous time slice. Substitute this into the second equation to obtain the current time slice. Taking a two-dimensional state vector as an example, if the actual state vector of a centrifugal pump in the current time slice is... The reference result output in step one is Deployment parameters Then, from the first equation, the current deviation vector is approximately If the deviation vector saved in the previous time slice is And the fault sensitivity weight row vector corresponding to the current device is taken as Trend weight row vector Then there is , The sum of the two terms yields the input to the logic mapping layer. Therefore, the fault prediction result for the current time slice is This is how it was obtained. This will be significantly higher than the result calculated solely based on the current deviation amplitude. This process reflects a common phenomenon in the early degradation of pump impellers or bearings: the deviation in a single time slice may not be large enough, but the direction of deviation increase in continuous time slices is relatively stable. Taking a CNC spindle as another example, within the same machining stage, the reference result given in step one already includes the expected vibration and current characteristics under normal cutting loads at that stage. This step further uses... Describe the proportion of the spindle's deviation relative to this reference result, and then use... Describing whether this deviation continues to increase, thus the slow vibration rise caused by early wear of the spindle bearing will appear earlier. The fault prediction results obtained in this way are not a simple response to the "abnormality of the original state value", but a continuous characterization of the "deviation from the digital twin reference state relative to the current operating condition and its evolution".

[0030] In the project deployment, The unified configuration is a single-value prediction quantity. The edge computing unit directly outputs a fault prediction intensity corresponding one-to-one with the current time slice for each time slice, which is then used as the basis for further evaluation in the next step. Because Based on the digital twin reference results output in step one Current time slice state vector Previous time slice deviation vector and the weight matrix corresponding to the device type It is calculated jointly, so it maintains the consistency of operating conditions and contains time evolution information, which can be directly used as input for generating evaluation results in the next step.

[0031] S3: Based on the fault prediction results on continuous time slices, construct the evaluation basis results including the current prediction level, trend normalization increment, and second-order difference curvature constraint.

[0032] Specifically, in step three, the fault prediction result has already been obtained in step two. Building upon this foundation, the goal is to transform predictions over consecutive time slices into assessment results that can reliably reflect equipment degradation trends. The result obtained in step two This is a consistent prediction of operating conditions calculated based on digital twin reference results. Essentially, it is a sequence signal corresponding one-to-one with each time slice. During the operation of industrial equipment, this sequence contains both actual equipment degradation information and local disturbances caused by factors such as control adjustments and load fluctuations. Therefore, this step focuses on the question of "how to extract stable trend information from the prediction sequence." Specifically, a sliding buffer arranged in chronological order is maintained in the edge computing unit to cache the current time slice. The previous time film And the previous time film These three quantities are all directly derived from the output of step two on continuous time slices, so the entire calculation process is entirely based on the output of the previous step.

[0033] First, construct the trend-normalized increment. This expression originates from the mathematical concepts of first-order difference and relative rate of change. First-order difference The relative rate of change is used to characterize the direction and magnitude of the change in the current prediction result relative to the previous time slice. The relative rate of change is normalized by the difference term to make changes at different amplitude levels comparable. In this application, the denominator of the relative rate of change is expanded from a single reference quantity to the sum of the amplitudes of the current and previous time slice predictions, and a stability term is introduced. This ensures that the denominator remains stable under any circumstances. This derivation is based on the following logic: the original relative rate of change is... ,Will Replace with information that includes both current and historical data. This can make the normalization scale smoother, and then add To avoid the denominator approaching zero. This leads to... ; in, The fault prediction result for the current time slice is calculated in step two and stored in the edge computing unit cache. This indicates the fault prediction result of the previous time slice, which is read from the cache in chronological order; These are stability parameters set during system deployment to ensure that the denominator is always positive. and This represents the magnitude of the corresponding forecast result, used to provide a normalized scale consistent with the current risk level; This represents the trend-normalized increment, and its value reflects the strength of the forecast's growth at the current risk level. Because... and The predicted values, obtained in step two through normalized bias, belong to the same scale, thus ensuring consistency in numerical scale between the left and right sides of the equation. Through this construction, when the equipment is in a stable state, and If the change is small, then Approaching zero; as equipment degradation gradually progresses, Persistently greater than ,but It is positive and increases with the growth rate; when instantaneous fluctuations occur, It is possible for positive and negative values ​​to alternate, while the denominator remains stable, making... It won't accumulate in the long term.

[0034] In obtaining Then, further construct the evaluation basis results. This expression is derived from the combination of proportional and rate-of-change terms in control theory, while introducing discrete second-order differences to suppress high-frequency fluctuations. The proportional term directly uses the current prediction result. The rate of change term adopts the aforementioned trend-normalized increment. Second-order difference term Derived from the description of sequence curvature in discrete signal processing, it is used to characterize whether a sequence has abrupt changes. The derivation process is as follows: Using only... At that time, the results are sensitive to instantaneous fluctuations; adding Then, it can reflect the trend; adding a second-order difference term and taking the absolute value can constrain abrupt changes, making the result smoother. Therefore, we obtain... ; in, This indicates the fault prediction result for the current time slice; This represents the trend normalization increment calculated from the previous formula; This is the trend enhancement coefficient, set according to the device type during system deployment, used to adjust the degree of influence of the trend on the results; This is the curvature suppression coefficient, used to control the suppression strength of the second-order difference term; and These represent the fault prediction results of the previous time slice and the time slice before that, respectively, both of which are read from the edge computing unit cache. It represents the magnitude of the discrete second-order difference, used to describe the degree of bending in the predicted sequence; This represents the non-negative truncation operation, which is implemented in the edge computing unit by comparing thresholds to ensure that the evaluation basis result is not less than zero. This indicates the assessment basis. The first term in the formula provides the current risk level, the second term reinforces the continuous growth trend, the third term suppresses sudden fluctuations, and the non-negative truncation operation further ensures that the assessment basis results always remain within the effective range that can be directly used in subsequent health mappings.

[0035] In actual computation, the edge computing unit executes the following process sequentially over time: reads the current time slice... and read from the cache and First, substitute into the first equation to get Then , , , Substituting into the second equation, we get For example, during the operation of a centrifugal pump, the prediction results for three consecutive time slices are as follows: , , ,Pick , , Then we get from the first equation The value is positive and relatively stable; substituting it into the second equation yields... Since the three-point sequence exhibits a smooth upward trend, the second-order difference term is relatively small. Compared to A slight increase, reflecting a continued downward trend. If a sudden fluctuation occurs at a certain time slice, for example, in a sequence... Then the second-order difference term increases significantly, thus suppressing the sudden fluctuation in the second equation, making It will not be affected by a single anomaly. The calculation process is executed in a time-slice loop within the edge computing unit, enabling the output of stable evaluation results in real time.

[0036] Through the above processing, step three transforms the fault prediction result sequence from step two into an evaluation basis result that simultaneously includes the current level, the trend of change, and the fluctuation constraint. The results can be directly used for subsequent dynamic health assessments, enabling the assessment process to be based on continuous evolutionary characteristics rather than single-point predictions, thus better reflecting the actual gradual degradation of industrial equipment.

[0037] S4: Generate a health baseline value by using the evaluation criteria results through exponential decay mapping, and dynamically correct the health baseline value by combining the changing trends of the evaluation criteria results in adjacent time slices, and output the dynamic health evaluation result.

[0038] Specifically, step four follows the evaluation results continuously output from step three. The assessment parameters, which already include information on the current level of degradation, the continuous growth trend, and the suppression of local fluctuations, are further mapped into dynamic health assessment results. In industrial equipment scenarios, health is not a direct transcription of degradation over a single time slice, but a dynamic quantity that needs to simultaneously express both the "current health status" and the "future evolution direction." Therefore, this step employs a two-stage calculation chain of "baseline mapping plus trend correction." The first stage uses the evaluation results as the basis for the calculation. The first part converts the current time slice's health baseline value into a second part, which then dynamically adjusts this baseline value based on the changing trends between adjacent time slices, resulting in the final output. It possesses both the ability to express state and the ability to express trends.

[0039] The health baseline value is constructed based on the exponential decay relationship commonly used in reliability engineering and life modeling. The original form of the exponential decay model comes from the exponential function in mathematics and the empirical rule in engineering that "the more deteriorated the condition, the faster the health declines." The classic notation is that a certain health value decreases exponentially with increasing degradation. Step three outputs the evaluation basis results. The degradation characteristic, already normalized and possessing stable physical semantics, can therefore be directly used as the independent variable in exponential decay. To accommodate the varying sensitivity of different devices to degradation, a scaling factor is introduced. The health baseline value is set during equipment commissioning based on the distribution differences between normal samples and known degraded samples, ensuring a usable slope for health changes across different equipment types. This yields the health baseline value for the current time slice. ; in, This indicates the evaluation results for the current time slice, which are calculated in step three and written to the cache. This represents the health scale adjustment coefficient, which is set according to the equipment type during the system deployment and debugging phase. For example, a larger value can be used for high-precision spindle equipment to enhance early degradation sensitivity, while a medium value can be used for pump and fan equipment to take into account stability. This indicates an exponential mapping operation, which is called through a numerical computing library in the edge computing unit; This represents the baseline health value for the current time slice. This formula extends the classic exponential decay relationship from general degradation modeling to evaluation based on results. Above: The degradation amount in the original model is output from step three. The replacement ensures that the health baseline is built directly on the output of the previous step, rather than on the original sensor state or instantaneous prediction. Because... It has been normalized. For a dimensionless quantity, the exponential mapping Similarly, it is a dimensionless health baseline value, suitable for use as a uniform scale for expressing health status in subsequent steps.

[0040] In obtaining Then, the final dynamic health assessment results are constructed. This part originates from the feedforward correction concept in discrete-time control, which introduces the change between adjacent time slices as a correction term based on the current state value to reflect the direction of system state evolution. Therefore, This indicates the change in the assessment results from the previous time slice to the current one, directly reflecting whether equipment degradation is accelerating, stabilizing, or slowing down. To ensure that the health results remain within the valid range, the trend correction uses a bounded fractional form: when... Compared to As the denominator increases, the health score further decreases; when and Close to or less than When, the denominator remains unchanged. The health level remains at the current baseline level. This yields the final dynamic health assessment result. ; in, This represents the baseline health value calculated using the previous formula; This indicates the evaluation results for the current time slice; The evaluation results of the previous time slice are read by the edge computing unit from the cache formed in step three in chronological order. This represents the trend adjustment coefficient, which is set based on continuously running samples when the device goes live, and is used to control the strength of the trend item's correction to the health status. This represents the positive trend cutoff amount, which is achieved by comparing thresholds in the edge computing unit, and only corrects the part where degradation continues to accelerate; This represents the dynamic assessment result of health status. There is a clear logical relationship between this formula and the previous formula: the previous formula first completes the static mapping from "deterioration amount to health status baseline value," and the latter formula then completes the "bounded correction of the baseline value by the accelerating deterioration trend." Because... and Coming from the same evaluation sequence and being a normalized quantity, As a dimensionless correction quantity, the denominator is always not less than Therefore, the final result is Always greater than zero and not exceeding .

[0041] In field applications, the edge computing unit executes a fixed process after each time slice. First, it reads the output of step three in the current time slice. Then read the previous time slice from the cache. Then, substitute the values ​​into the first formula to obtain the base health value. Then, substituting into the second equation, we obtain the final dynamic health assessment result. Finally, the current Write to the cache for the next time slice as a new Use. Taking a centrifugal pump as an example, if the evaluation result of the current time slice is... The previous video was Debugging settings , First, calculate the health baseline value. The value is approximately Then calculate the denominator. Therefore, the final dynamic health assessment result is approximately This result indicates that the pump has already undergone some degree of degradation, and because the assessment results continue to increase compared to the previous time slice, its health relative to the baseline value has further decreased. If in another time slice... and Then the denominator term is The final health result is the same as the baseline value, indicating that the current degradation level is stable. If the CNC spindle recovers briefly after tool change, it forms... , Under the same parameters, first obtain Then we get the denominator term. ,final This indicates that the spindle's current health level remains at the current baseline. Following this calculation chain, the health result naturally exhibits a stable, gradual, or accelerated decline over continuous time slices, consistent with the actual degradation and evolution process of industrial equipment.

[0042] During engineering commissioning, continuous operation records collected on-site can be used to... and Calibration is performed. This is done by collecting a set of data output from step three during the equipment's normal operation, early degradation, and significant degradation phases. The sequence is used, and the equipment status in maintenance or repair records is compared. For each type of equipment, the status is first determined based on the normal stage and the early degradation stage. Distribution interval determined This ensures that the baseline health score remains high within the normal range and shows a significant decline in the early deterioration range; then, based on the continuous time slices... growth rate adjustment This allows the health curve to decline promptly when degradation accelerates and to remain stable during short-term disturbances. The debug log typically saves data for each time slice. , , and The data includes numerical values, corresponding operating condition numbers, and maintenance time points. These records can verify that in typical scenarios such as fan bearing degradation, pump impeller wear, and early damage to the main shaft bearing... All of them exhibit a sustained downward trend early before the fault spreads, and the rate of decline is consistent with the rate of degradation. Thus, step four not only completes the assessment based on the results... To the dynamic assessment results of health status The mathematical mapping also provides a complete operation path for parameter configuration, time slice calculation, result output and debugging calibration at the field implementation level.

[0043] In one or more embodiments, such as Figure 2 As shown, a digital twin-based equipment failure prediction and health assessment system is disclosed, the system comprising: The data acquisition module is used to acquire equipment status data and operating condition identification data of industrial equipment, input the equipment status data and operating condition identification data into a pre-deployed digital twin model, and generate a digital twin reference result that strictly corresponds to the current operating segment. The fault prediction module is used to retrieve the actual state vector that is synchronously saved with the digital twin reference result, calculate the normalized deviation vector of the operating condition based on the actual state vector and the digital twin reference result, and generate the fault prediction result by combining the evolution relationship of adjacent time slices of the normalized deviation vector of the operating condition. The evaluation basis module is used to construct an evaluation basis result that includes the current prediction level, trend normalization increment, and second-order difference curvature constraint based on the fault prediction results on continuous time slices. The dynamic evaluation module is used to generate a health baseline value by using the evaluation basis results through exponential decay mapping, and dynamically correct the health baseline value by combining the changing trend of the evaluation basis results of adjacent time slices, and output the dynamic evaluation result of health.

[0044] It is worth noting that the specific workflow of the equipment fault prediction and health assessment system based on digital twins provided in this embodiment of the invention is the same as that of the equipment fault prediction and health assessment method based on digital twins described in the above embodiments, and will not be repeated here.

[0045] This invention also provides a device for device fault prediction and health assessment based on digital twins, including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor. When the processor executes the computer program, it implements the steps described in the above embodiments of the device fault prediction and health assessment method based on digital twins. Figure 1 The steps S1 to S4 described above; or, when the processor executes the computer program, it implements the functions of each module in the above system embodiments.

[0046] For example, the computer program may be divided into one or more modules, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which describe the execution process of the computer program in the digital twin-based device fault prediction and health assessment device.

[0047] The device fault prediction and health assessment device based on digital twins can be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. This device may include, but is not limited to, processors and memory. Those skilled in the art will understand that the device may also include input / output devices, network access devices, buses, etc.

[0048] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the digital twin-based equipment fault prediction and health assessment device, connecting all parts of the device via various interfaces and lines.

[0049] The memory can be used to store the computer programs and / or modules. The processor, by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory, realizes various functions of the digital twin-based device fault prediction and health assessment device. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store data created according to the operation of the controller, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital card (SD card), flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0050] The module integrated into the digital twin-based device fault prediction and health assessment device, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, a recording medium, a USB flash drive, a portable hard drive, a magnetic disk, an optical disk, a computer memory, a read-only memory (ROM), a random access memory (RAM), an electrical carrier signal, a telecommunication signal, and a software distribution medium, etc.

[0051] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory (ROM), or random access memory (RAM), etc.

[0052] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method for equipment fault prediction and health assessment based on digital twins, characterized in that, The method includes: Acquire equipment status data and operating condition identification data of industrial equipment, input the equipment status data and operating condition identification data into a pre-deployed digital twin model, and generate a digital twin reference result that strictly corresponds to the current operating segment; The actual state vector, which is synchronously stored with the digital twin reference result, is retrieved in the edge computing unit. Based on the actual state vector and the digital twin reference result, a normalized operating condition deviation vector is calculated. Combining the evolution relationship of adjacent time slices of the normalized operating condition deviation vector, a fault prediction result is generated. The step of calculating the normalized operating condition deviation vector based on the actual state vector and the digital twin reference result specifically involves: calculating the difference between the actual state vector and the digital twin reference result, and dividing the difference by the denominator, which includes the sum of the absolute values ​​of the preset stability term and the digital twin reference result, to obtain the normalized operating condition deviation vector. Based on the fault prediction results on continuous time slices, an evaluation basis result is constructed, which includes the current prediction level, trend normalization increment, and second-order difference curvature constraint. The specific steps of constructing the evaluation basis result include: calculating the trend normalization increment between the fault prediction result of the current time slice and the fault prediction result of the previous time slice; calculating the absolute value of the second-order difference between the fault prediction results of the current time slice, the previous time slice, and the time slice before that; weighting and combining the fault prediction result of the current time slice, the trend normalization increment, and the absolute value of the second-order difference, and then performing non-negative truncation to generate the evaluation basis result. The evaluation criteria results are used to generate a health baseline value through exponential decay mapping. The health baseline value is then dynamically corrected by combining the changing trends of the evaluation criteria results in adjacent time slices, and the dynamic health evaluation result is output.

2. The method for equipment fault prediction and health assessment based on digital twins according to claim 1, characterized in that, In the step of inputting the equipment status data and operating condition identification data into a pre-deployed digital twin model to generate a digital twin reference result that strictly corresponds to the current operating segment, the digital twin model includes an operating condition mapping layer and a status response layer; the operating condition mapping layer is used to generate current operating condition parameters based on the operating condition identification data, and the status response layer is used to generate a digital twin reference result based on the current operating condition parameters and the equipment status data.

3. The method for equipment fault prediction and health assessment based on digital twins according to claim 1, characterized in that, The step of generating fault prediction results by combining the evolution relationship of adjacent time slices of the normalized deviation vector of operating conditions specifically includes: weighted summation of the normalized deviation vector of the current time slice and the fault-sensitive weight row vector; simultaneously calculating the trend difference between the normalized deviation vector of the current time slice and the previous time slice and combining it with the trend weight row vector; and outputting the above two calculation results as a single-value fault prediction result through the logical mapping layer. The fault-sensitive weight row vector originates from the offline training results before system deployment; the trend weight row vector is used to control the contribution of the changing trends of each deviation component in the prediction, and is configured according to the equipment type during deployment; the offline training uses normalized deviation samples of operating conditions during the normal operation phase and the confirmed early degradation phase as input, with the corresponding degradation label as the target, and is obtained through constrained linear regression or logistic regression.

4. The method for equipment fault prediction and health assessment based on digital twins according to claim 1, characterized in that, The step of generating a health baseline value by exponential decay mapping of the evaluation basis result specifically involves multiplying the evaluation basis result by a health scale adjustment coefficient and taking a negative exponent to generate the health baseline value for the current time slice.

5. The method for equipment fault prediction and health assessment based on digital twins according to claim 1, characterized in that, The step of dynamically correcting the health baseline value by combining the changing trend of the evaluation results of adjacent time slices specifically involves: calculating the difference between the evaluation results of the current time slice and the evaluation results of the previous time slice. If the difference is greater than zero, the product of the difference and the trend adjustment coefficient is used as the denominator correction term to perform bounded correction on the health baseline value, generating the final dynamic health assessment result.

6. The method for equipment fault prediction and health assessment based on digital twins according to claim 1, characterized in that, The equipment status data includes at least one of vibration characteristics, current characteristics, temperature characteristics, and pressure characteristics; the operating condition identification data includes at least one of speed range number, load range number, process section number, or recipe number.

7. A digital twin-based equipment fault prediction and health assessment system, characterized in that, The system includes: The data acquisition module is used to acquire equipment status data and operating condition identification data of industrial equipment, input the equipment status data and operating condition identification data into a pre-deployed digital twin model, and generate a digital twin reference result that strictly corresponds to the current operating segment. The fault prediction module is used to retrieve the actual state vector that is synchronously saved with the digital twin reference result, calculate the normalized deviation vector of the operating condition based on the actual state vector and the digital twin reference result, and generate the fault prediction result by combining the evolution relationship of adjacent time slices of the normalized deviation vector of the operating condition; the step of calculating the normalized deviation vector of the operating condition based on the actual state vector and the digital twin reference result specifically includes: calculating the difference between the actual state vector and the digital twin reference result, and dividing the difference by the denominator containing the sum of the absolute values ​​of the preset stability term and the digital twin reference result to obtain the normalized deviation vector of the operating condition; The evaluation basis module is used to construct an evaluation basis result based on the fault prediction results over continuous time slices, including the current prediction level, trend normalization increment, and second-order difference curvature constraint. The step of constructing the evaluation basis result specifically includes: calculating the trend normalization increment between the fault prediction result of the current time slice and the fault prediction result of the previous time slice; calculating the absolute value of the second-order difference between the fault prediction results of the current time slice, the previous time slice, and the time slice before that; and weighting and combining the fault prediction result of the current time slice, the trend normalization increment, and the absolute value of the second-order difference, and then truncating them non-negatively to generate the evaluation basis result. The dynamic evaluation module is used to generate a health baseline value by using the evaluation basis results through exponential decay mapping, and dynamically correct the health baseline value by combining the changing trend of the evaluation basis results of adjacent time slices, and output the dynamic evaluation result of health.

Images