A device asset intelligent digital management method and system

Abstract

The application discloses a kind of equipment asset intelligent digital management method and system, it is related to equipment management technical field, the method includes: using sensor cluster to collect the multimodal time series data stream that physical equipment sends out, and it is converted into high-dimensional feature vector;Using high-dimensional feature vector deduces the internal state variable of current time equipment, generates equipment health state vector after combination;Analysis equipment health state vector under time series arrangement, generates health degree degradation trajectory;Input health degree degradation trajectory to decision maker, generates the optimal maintenance action sequence for equipment;The maintenance action sequence generated is executed, and the decision maker is continuously optimized according to the equipment state after execution.The application breaks the traditional equipment management information island, can reduce unplanned downtime, reduce maintenance cost, improve equipment utilization and management efficiency, realize the change from passive maintenance to predictive management.

Description

Technical Field

[0001] This invention relates to the field of equipment management technology, and in particular to an intelligent digital management method and system for equipment assets. Background Technology

[0002] In modern industrial production, infrastructure operation, and the daily operations of large institutions (such as hospitals and universities), equipment assets are a core production factor, and their management efficiency directly affects operating costs, production safety, and economic benefits. Traditional equipment asset management models primarily rely on manual ledgers, regular inspections, and planned maintenance, but their limitations are becoming increasingly apparent, mainly in the following aspects:

[0003] First, the phenomenon of information silos is severe, making it difficult to extract the value of data. Enterprises' equipment asset information is typically scattered across different management departments and systems, such as procurement information in ERP systems, maintenance records in CMMS systems, and real-time operational data in SCADA systems. These systems lack effective data connectivity and integration, forming "information silos." Managers struggle to obtain a complete and unified view of the entire equipment lifecycle, hindering their ability to conduct holistic data analysis and make optimization decisions.

[0004] Second, maintenance strategies are reactive and outdated, leading to high costs for unplanned downtime. Traditional maintenance models are mostly breakdown maintenance or fixed planned preventive maintenance. The former usually intervenes only after equipment failure, resulting in unplanned downtime and causing huge production losses and safety risks; the latter is based on fixed time or operating cycles, which may lead to "over-maintenance," wasting resources, or failure to detect potential faults in time, resulting in under-maintenance. Neither of these models can accurately predict the health status of the equipment.

[0005] Third, the management process is highly dependent on manual labor, resulting in low efficiency and a high risk of errors. From asset inventory to status recording, from work order issuance to maintenance reports, a large amount of work relies on on-site inspections by managers and paper or Excel spreadsheet recording. This approach is not only inefficient and costly in terms of labor, but also highly susceptible to data recording errors, information transmission delays or omissions due to human factors, significantly compromising the accuracy and timeliness of management data.

[0006] Fourth, the unknowable status of assets makes full lifecycle management difficult. For a large number of widely distributed equipment assets, managers struggle to obtain real-time and accurate information on their location, current operating status, utilization rate, performance degradation, and other dynamic data. This leads to widespread asset idleness and waste, making it impossible to make optimal allocation, renewal, or scrapping decisions based on real-time data, thus hindering the achievement of asset preservation and appreciation goals.

[0007] With the maturation and widespread adoption of next-generation information technologies such as the Internet of Things (IoT), big data, artificial intelligence (AI), and cloud computing, technological solutions to the aforementioned problems have become possible. Solutions attempting to utilize these technologies for equipment management have emerged within the industry, such as collecting equipment operation data by adding sensors or using RFID and QR codes for asset identification. However, existing solutions often merely "digitize" traditional management processes without truly achieving "intelligentization." The problems lie in: limited data collection dimensions and a lack of multi-source data fusion and analysis; weak data analysis capabilities, mostly limited to data visualization and simple alarms, lacking deep learning and predictive analytics capabilities; and rigid system architectures, making it difficult to flexibly expand and adapt to complex and ever-changing management scenarios.

[0008] Therefore, there is an urgent need in this field for an intelligent digital management method for equipment assets that deeply integrates advanced information technology, breaks down information silos, enables data-driven decision-making, and has self-learning and adaptive capabilities, so as to fundamentally improve the reliability, utilization and management efficiency of equipment assets and reduce the total life cycle cost. Summary of the Invention

[0009] This invention provides an intelligent digital management method for equipment assets, comprising:

[0010] Step 1: Use a sensor cluster to collect multimodal time-series data streams emitted by physical devices and convert them into high-dimensional feature vectors;

[0011] Step 2: Use high-dimensional feature vectors to deduce the internal state variables of the device at the current moment, and combine them to generate the device health state vector;

[0012] Step 3: Analyze the device health status vector under the time sequence arrangement to generate the health degradation trajectory;

[0013] Step 4: Input the health degradation trajectory into the decision-maker to generate the optimal maintenance action sequence for the equipment;

[0014] Step 5: Execute the generated maintenance action sequence and continuously optimize the decision-maker based on the equipment status after execution.

[0015] The intelligent digital management method for equipment assets described above involves transforming multimodal time-series data streams into high-dimensional feature vectors, specifically through the following sub-steps:

[0016] Perform time alignment and preprocessing on the data stream acquired by the sensor cluster;

[0017] The preprocessed data stream is input into a pre-trained dedicated feature extraction neural network to extract the vibration feature vector, temperature feature vector and acoustic feature vector of the equipment;

[0018] The extracted modal feature vectors are concatenated to generate a high-dimensional feature vector.

[0019] The intelligent digital management method for equipment assets described above, which uses high-dimensional feature vectors to deduce the internal state variables of the equipment at the current moment and combines them to generate an equipment health status vector, is specifically divided into the following sub-steps:

[0020] A baseline health attractor is constructed based on multimodal time-series data from health devices;

[0021] Calculate the deviation energy level vector between the short-term state trajectory formed by the current high-dimensional eigenvectors and the baseline healthy attractor;

[0022] Generate the current device health status vector based on the calculated deviation energy level vector.

[0023] The intelligent digital management method for equipment assets described above, which generates a health degradation trajectory, specifically comprises the following sub-steps:

[0024] Adaptive enhancement of the device health status vector based on contextual data of the device's operating environment;

[0025] Calculate the degradation factor of each dimension of the enhanced device health status vector;

[0026] The health status of the device at each time step t is calculated based on the degradation factors of each dimension, and then organized into the original health status sequence.

[0027] Input the original health sequence into the GPR model and output the complete health degradation trajectory.

[0028] The intelligent digital management method for equipment assets described above specifically includes the following sub-steps in its decision-maker training process:

[0029] Collect historical equipment operation data to build a training dataset;

[0030] Freeze the policy network of the decision-maker and initialize its evaluation network based on the training dataset;

[0031] Unfreeze the policy network of the decision-maker, combine it with the evaluation network to calculate the target loss of the decision-maker in real time, and fine-tune the decision-maker based on the principle of minimum loss until the model converges.

[0032] The present invention also provides an intelligent digital management system for equipment assets, comprising: a data flow conversion module, a health status vector generation module, a health degradation trajectory generation module, a maintenance action sequence generation module, and a maintenance action sequence execution module;

[0033] The data stream conversion module is used to collect multimodal time-series data streams emitted by physical devices using a sensor cluster and convert them into high-dimensional feature vectors.

[0034] The health status vector generation module is used to deduce the internal state variables of the device at the current moment using high-dimensional feature vectors, and then combine them to generate the device health status vector.

[0035] The health degradation trajectory generation module is used to analyze the device health status vector under time sequence arrangement and generate health degradation trajectory;

[0036] The maintenance action sequence generation module is used to input the health degradation trajectory into the decision-maker and generate the optimal maintenance action sequence for the equipment.

[0037] The maintenance action sequence execution module is used to execute the generated maintenance action sequence and continuously optimize the decision-maker based on the equipment status after execution.

[0038] The beneficial effects achieved by this invention are as follows: it breaks down the information silos of traditional equipment management, reduces unplanned downtime, lowers maintenance costs, improves equipment utilization and management efficiency, and realizes the transformation from passive maintenance to predictive management. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0040] Figure 1 This is a flowchart of an intelligent digital management method for equipment assets provided in Embodiment 1 of this application. Detailed Implementation

[0041] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0042] Example 1

[0043] like Figure 1 As shown, Embodiment 1 of this application provides an intelligent digital management method for equipment assets, including:

[0044] Step S110: Use a sensor cluster to collect multimodal time-series data streams emitted by physical devices and convert them into high-dimensional feature vectors;

[0045] This step is the foundational layer for data perception and feature extraction, and its specific implementation process is as follows:

[0046] Step S111: Perform time alignment and preprocessing on the data stream acquired by the sensor cluster;

[0047] The sensor cluster includes at least: a vibration sensor for collecting the vibration spectrum of the equipment in the X, Y, and Z axes; a temperature sensor for collecting the surface temperature distribution sequence of key components of the equipment; an acoustic sensor for collecting noise signals during equipment operation to generate acoustic signatures; and an environmental sensor (including temperature and humidity sensors and dust sensors) for collecting contextual data of the operating environment of the equipment, i.e., the temperature, humidity, and dust levels of the operating environment.

[0048] Since each sensor operates independently and has a different sampling rate, the resulting raw data stream is asynchronous. Therefore, it is necessary to use an IoT gateway to add a high-precision unified timestamp to each frame of data, and use a time series alignment algorithm (such as dynamic time warping (DTW) or linear interpolation) to align the data of all modes onto the same time axis to form a multi-channel synchronous data block. Then, the synchronous data block is preprocessed to obtain a standardized data stream.

[0049] Step S112: Input the preprocessed data stream into the pre-trained dedicated feature extraction neural network to extract the vibration feature vector, temperature feature vector and acoustic feature vector of the device;

[0050] The dedicated feature extraction neural network comprises three extraction branches: a vibration feature extraction branch, which uses a one-dimensional convolutional neural network to extract impact features and sideband features from the vibration spectrum, outputting a fixed-length vibration feature vector; a temperature feature extraction branch, which uses a two-dimensional convolutional neural network to extract the spatial distribution features of the temperature field from the temperature distribution sequence, outputting a temperature feature vector; and a voiceprint feature extraction branch, which converts the voiceprint spectrum into a time-spectrum image, and then uses another two-dimensional convolutional neural network to extract acoustic features, outputting an acoustic feature vector. Many pre-trained models exist for extracting vibration spectrum, temperature field, and voiceprint spectrum features; no limitation is imposed here.

[0051] Step S113: Concatenate the extracted modal feature vectors to generate a high-dimensional feature vector;

[0052] By splicing the modal feature vectors extracted by the dedicated feature extraction neural network, a unified high-dimensional feature vector that can comprehensively represent the current instantaneous state of the device is obtained.

[0053] Step S120: Use high-dimensional feature vectors to deduce the internal state variables of the device at the current moment, and combine them to generate the device health state vector;

[0054] The aforementioned deduction method refers to abstracting the device as a dynamic system, whose health state evolution forms a trajectory in the state space. By reconstructing the system's "health attractor" and calculating the "deviation energy level" of the real-time feature vector relative to the attractor, the internal state variables are accurately deduced. Specifically, it consists of the following sub-steps:

[0055] Step S121: Construct a baseline health attractor based on multimodal time-series data from health devices;

[0056] With the equipment in a brand-new or healthy state, a large amount of multimodal time-series data is collected, and a high-dimensional feature vector set is obtained through step S110. Then, the high-dimensional feature vector is reduced to a low-dimensional state space using an autoencoder. In this space, the health state data points will cluster to form a compact region, which is called the baseline health attractor, denoted as M.

[0057] Step S122: Calculate the deviation energy level vector between the short-term state trajectory formed by the current high-dimensional feature vectors and the baseline healthy attractor;

[0058] The high-dimensional feature vector at the current moment Projecting onto the aforementioned low-dimensional state space yields the projection point. By introducing a sliding window mechanism, the sequence of projection points within the most recent time period can be obtained. This refers to the short-term state trajectory of the equipment.

[0059] The deviation energy level vector The calculation formula is expressed as:

[0060] in It is the projection point of the high-dimensional feature vector at the current moment. It is any point in the baseline health attractor M. It is a short-term state trajectory The i-th, i+1-th, and i-1-th trajectory points in the array, where i takes values ​​from 2 to n-1, and n is... The number of trajectory points in the data. The resilience factor is used to quantify the natural resilience of a device, calibrated using historical data. As a baseline, the health attractor M is in The closest point.

[0061] Step S123: Generate the current device health status vector based on the calculated deviation energy level vector;

[0062] First, each component in the calculated deviation energy level vector is normalized. Then, an adaptive weighting mechanism is introduced, assigning corresponding weight coefficients based on the relative importance of each component. Finally, the weighted deviation energy level components are combined to form the device health state vector at this moment. ,in These represent the individual components in the deviation energy level vector.

[0063] Step S130: Analyze the device health status vector under the time sequence arrangement and generate the health degradation trajectory;

[0064] Based on the time-series equipment health status vector output in step S120, and combined with the operating context data collected in step S110, the nonlinear dynamic law of equipment health degradation is mined to generate a complete health degradation trajectory of "historical smooth trajectory + future predicted trend". This is specifically divided into the following sub-steps:

[0065] Step S131: Adaptively enhance the device health status vector based on the context data of the device's operating environment;

[0066] For the device health status vector at each time step t Take the front and back Each effective time step constitutes a local neighborhood. Next, extract the context data of the device's operating environment at the corresponding time step, convert it into a vector, and add it to the local neighborhood. Then, substitute the data from the local neighborhood into the formula: In this process, the enhanced health state vector is obtained. ,in This is a sensitivity adjustment parameter, assigned a value based on the importance placed on the health of the local neighborhood. j is the time step index within the local neighborhood, and m is the total number of samples within the local neighborhood. These are the device health status vector and the operating environment vector at the j-th time step within the local neighborhood, respectively. It is the working environment vector at the current time step t. It is the maximum difference value of the global operating conditions. It is the cosine similarity function.

[0067] Step S132: Calculate the degradation factor of each dimension of the enhanced device health status vector;

[0068] The formula for calculating the degradation factor is expressed as follows:

[0069] in Let d represent the degradation factor of the d-th dimension of the device health state vector at time step t. , These are the d-th dimension components of the device health status vector at time step t and time step t-1, respectively. It is the time interval between time steps t and t-1. It is the time-series sequence of the d-th dimension component of the device health status vector. It is a time-series vector sequence of operating conditions and environment. It is the covariance operation function. It is the standard deviation operation function. This refers to the dimension index during the summation process, where D is the total dimension of the device health status vector. The meanings of the parameters of dimension 1 are the same as those of the parameters of dimension d, so they will not be repeated here.

[0070] Step S133: Calculate the health status of the device at each time step t based on the degradation factors of each dimension, and organize it into the original health status sequence;

[0071] Device health at time step t The calculation formula is expressed as follows:

[0072]

[0073] in These are the adjustable scaling parameter and the nonlinear coefficient, respectively. It is the maximum allowable value of the d-th dimension of the device health status vector under the device health status, which is calibrated by the device's historical detection data.

[0074] Step S134: Input the original health sequence into the GPR model and output the complete health degradation trajectory;

[0075] The GPR (Gaussian Process Regression) model is used to generate a smooth historical health trajectory from the original health sequence, while simultaneously predicting future degradation trends, i.e., future health degradation trajectories. Concatenating these two trajectories forms a complete health degradation trajectory, thereby achieving a shift from passive maintenance to predictive maintenance. The kernel function used in the GPR model is represented as follows: ,in For two different time steps index, For time core, This is used to ensure the temporal continuity of health tracking. For operating condition verification, It is used to correct the interference of operating condition fluctuations on the trajectory. These are adjustable time sensitivity correction parameters and operating condition sensitivity correction parameters.

[0076] Step S140: Input the health degradation trajectory into the decision-maker to generate the optimal maintenance action sequence for the equipment;

[0077] The decision-maker receives the health degradation trajectory as input and generates the optimal maintenance action sequence. Its training process is specifically divided into the following sub-steps:

[0078] Step S141: Collect historical equipment operation data and construct a training dataset;

[0079] Collect historical equipment operation data, including health sequence, maintenance action records and post-maintenance status feedback. Divide the data into multiple segments according to the maintenance cycle. Each segment contains a health degradation trajectory, the sequence of executed maintenance actions and their corresponding health change results, forming a training dataset.

[0080] Step S142: Freeze the policy network of the decision-maker and initialize its evaluation network based on the training dataset;

[0081] The decision-maker includes a policy network and an evaluation network. The policy network is used to generate a sequence of maintenance actions based on the input health degradation trajectory, and the evaluation network is used to predict the health trend after executing the maintenance sequence.

[0082] In this stage, only the training dataset is used to supervise the evaluation network so that it can accurately predict the future health trend of the device given a health trajectory and a sequence of maintenance actions.

[0083] Step S143: Unfreeze the policy network of the decision-maker, calculate the target loss of the decision-maker in real time in conjunction with the evaluation network, and fine-tune the decision-maker based on the principle of minimum loss until the model converges;

[0084] Unfreeze the policy network of the decision maker, and define the objective loss function as: Where b is the training sample index, B is the number of samples in the current training batch, t is the time step index, and T is the length of the health trajectory predicted by the evaluation network. It is the discount factor for time step t. It is the maintenance action sequence output by the policy network for the b-th training sample. Indicating targeting The health status at time step t in the predicted health status trajectory. It's about the health status of t a while ago. This represents the device health at the last trajectory point of the health degradation trajectory in the b-th training sample. To maintain cost weighting, Indicates the maintenance action sequence The total cost is calculated based on historical maintenance records. The policy network and evaluation network are alternately optimized using the gradient descent algorithm to minimize the objective loss until the model converges.

[0085] Step S150: Execute the generated maintenance action sequence and continuously optimize the decision-maker based on the equipment status after execution;

[0086] First, the maintenance action sequence generated by the decision-maker is parsed into a standardized electronic work order containing specific operating instructions, safety specifications, required spare parts, and estimated working hours. Then, it is automatically sent to the corresponding maintenance personnel's mobile terminals or automated maintenance equipment through the system interface. After execution, the decision-maker is continuously optimized based on the multimodal time-series data stream fed back by the sensor cluster.

[0087] The current health status of the device is calculated based on the multimodal time-series data stream fed back by the sensor cluster;

[0088] If the health status exceeds the predicted health status, then the decision weight of each type of maintenance action in the current maintenance action sequence will be increased.

[0089] If the health status does not reach the predicted health status, the prediction bias of the evaluation network is corrected.

[0090] Example 2

[0091] Embodiment 2 of this application provides an intelligent digital management system for equipment assets, including: a data flow conversion module, a health status vector generation module, a health degradation trajectory generation module, a maintenance action sequence generation module, and a maintenance action sequence execution module;

[0092] (1) Data stream conversion module, which is used to collect multimodal time-series data streams emitted by physical devices using sensor clusters and convert them into high-dimensional feature vectors; specifically including: data stream preprocessing submodule, feature vector extraction submodule, and high-dimensional feature vector generation submodule;

[0093] 1. Data stream preprocessing submodule, used to perform time alignment and preprocessing on the data streams collected by the sensor cluster;

[0094] 2. Feature vector extraction submodule, which is used to input the preprocessed data stream into the pre-trained dedicated feature extraction neural network to extract the vibration feature vector, temperature feature vector and acoustic feature vector of the equipment;

[0095] 3. High-dimensional feature vector generation submodule, used to concatenate the extracted modal feature vectors to generate high-dimensional feature vectors.

[0096] (2) Health status vector generation module, which is used to deduce the internal state variables of the device at the current moment using high-dimensional feature vectors, and generate the device health status vector after combination.

[0097] (3) Health degradation trajectory generation module, used to analyze the device health status vector under time sequence arrangement and generate health degradation trajectory; specifically including: feature enhancement submodule, health sequence generation submodule, and health degradation trajectory improvement submodule;

[0098] 1. Feature enhancement submodule, used to adaptively enhance the device health status vector based on contextual data of the device's operating environment;

[0099] 2. The health sequence generation submodule is used to calculate the health of the device at each time step t based on the degradation factors of each dimension of the device health status vector, and organize it into the original health sequence.

[0100] 3. The Health Degradation Trajectory Improvement Submodule is used to input the original health sequence into the GPR model and output the complete health degradation trajectory.

[0101] (4) Maintenance action sequence generation module, used to input the health degradation trajectory to the decision maker and generate the optimal maintenance action sequence for the equipment.

[0102] (5) Maintenance action sequence execution module, used to execute the generated maintenance action sequence and continuously optimize the decision-maker based on the equipment status after execution.

[0103] Corresponding to the above embodiments, the present invention provides a computer storage medium, including: at least one memory and at least one processor;

[0104] The memory is used to store one or more program instructions;

[0105] A processor is used to run one or more program instructions to execute an intelligent digital management method for equipment assets.

[0106] Corresponding to the above embodiments, this embodiment of the invention provides a computer-readable storage medium containing one or more program instructions, which are executed by a processor to provide an intelligent digital management method for equipment assets.

[0107] The embodiments disclosed in this invention provide a computer-readable storage medium storing computer program instructions, which, when executed on a computer, cause the computer to perform the aforementioned intelligent digital management method for equipment assets.

[0108] In this embodiment of the invention, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0109] The various methods, steps, and logic diagrams disclosed in the embodiments of this invention can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The processor reads information from the storage medium and, in conjunction with its hardware, completes the steps of the above methods.

[0110] The storage medium can be memory, such as volatile memory or non-volatile memory, or may include both volatile and non-volatile memory.

[0111] Among them, non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory.

[0112] Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (Synchlink DRAM, SLDRAM), and direct memory bus RAM (DRRAM).

[0113] The storage media described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memory.

[0114] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this invention can be implemented using a combination of hardware and software. When applied as software, the corresponding functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of computer programs from one place to another. Storage media can be any available medium that can be accessed by a general-purpose or special-purpose computer.

[0115] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for intelligent digital management of equipment assets, characterized in that, include: Step 1: Use a sensor cluster to collect multimodal time-series data streams emitted by physical devices and convert them into high-dimensional feature vectors; The sensor cluster includes at least: a vibration sensor for acquiring the vibration spectrum of the device along the X, Y, and Z axes; a temperature sensor for acquiring the surface temperature distribution sequence of key components of the device; and an acoustic sensor for acquiring noise signals during device operation to generate acoustic signatures. Step 2: Use high-dimensional feature vectors to deduce the internal state variables of the device at the current moment, and combine them to generate the device health state vector. This is divided into the following sub-steps: Constructing a baseline health attractor based on multimodal time-series data of health devices: The multimodal time-series data of the device's health status is processed into a set of high-dimensional feature vectors, and then an autoencoder is used to reduce the dimensionality of the obtained high-dimensional feature vectors to a low-dimensional state space. In this space, the health status data points will cluster to form a compact region, which is called the baseline health attractor. Calculate the deviation energy level vector between the short-term state trajectory formed by the current high-dimensional feature vectors and the baseline healthy attractor: [This involves] converting the current high-dimensional feature vectors... Projecting onto the aforementioned low-dimensional state space yields the projection point. By introducing a sliding window mechanism, the sequence of projection points within the most recent time period can be obtained. That is, the short-term state trajectory of the device; the deviation energy level vector The calculation formula is expressed as: ,in It is the projection point of the high-dimensional feature vector at the current moment. It is any point in the baseline health attractor M. It is a short-term state trajectory The i-th, i+1-th, and i-1-th trajectory points in the array, where i takes values ​​from 2 to n-1, and n is... The number of trajectory points in the data. The resilience factor is used to quantify the natural resilience of a device, calibrated using historical data. As a baseline, the health attractor M is in The nearest point; The device health status vector at this moment is generated based on the calculated deviation energy level vector: First, each component in the calculated deviation energy level vector is normalized. Then, an adaptive weighting mechanism is introduced to assign corresponding weight coefficients according to the relative importance of each component. Finally, the weighted deviation energy level components are combined to form the device health status vector at this moment. Step 3: Analyze the device health status vector under the time sequence arrangement to generate the health degradation trajectory, which is divided into the following sub-steps: Adaptive enhancement of the device health status vector based on contextual data of the device's operating environment: For the device health status vector at each time step t Take the front and back Each effective time step constitutes a local neighborhood. Next, extract the context data of the device's operating environment at the corresponding time step, convert it into a vector, and add it to the local neighborhood. Then, substitute the data from the local neighborhood into the formula: In this process, the enhanced health state vector is obtained. ,in Here, j is the sensitivity adjustment parameter, j is the time step index within the local neighborhood, and m is the total number of samples within the local neighborhood. These are the device health status vector and the operating environment vector at the j-th time step within the local neighborhood, respectively. It is the working environment vector at the current time step t. It is the maximum difference value of the global operating conditions. It is the cosine similarity function; Calculate the degradation factor for each dimension of the enhanced device health status vector; the formula for calculating the degradation factor is expressed as: ,in Let d represent the degradation factor of the d-th dimension of the device health status vector at time step t. , These are the d-th dimension components of the device health status vector at time step t and time step t-1, respectively. It is the time interval between time steps t and t-1. It is the time-series sequence of the d-th dimension component of the device health status vector. It is a time-series vector sequence of operating conditions and environment. It is the covariance operation function. It is the standard deviation operation function. This refers to the dimension index during the summation process, where D is the total dimension of the device health status vector. The meanings of the parameters in dimension 1 are consistent with those of the parameters in dimension d. The health status of the device at each time step t is calculated based on the degradation factors of each dimension, and then compiled into the original health status sequence; the health status of the device at time step t. The calculation formula is expressed as follows: ,in These are the adjustable scaling parameter and the nonlinear coefficient, respectively. It is the maximum allowable value of the d-th dimension of the device health state vector under the device health state; Input the original health status sequence into the GPR model, and output the complete health status degradation trajectory; Step 4: Input the health degradation trajectory into the decision-maker to generate the optimal maintenance action sequence for the equipment; the training process of the decision-maker is specifically divided into the following sub-steps: Collect historical equipment operation data to construct a training dataset; the historical equipment operation data includes health sequence, maintenance action record and post-maintenance status feedback. The data is divided into multiple segments according to the maintenance cycle. Each segment contains a health degradation trajectory, the sequence of executed maintenance actions and its corresponding health change results to form a training dataset. The policy network of the decision-maker is frozen, and its evaluation network is initialized based on the training dataset. The decision-maker includes a policy network and an evaluation network. The policy network is used to generate a sequence of maintenance actions based on the input health degradation trajectory, and the evaluation network is used to predict the health trend after the maintenance sequence is executed. Unfreeze the policy network of the decision-maker, combine it with the evaluation network to calculate the target loss of the decision-maker in real time, and fine-tune the decision-maker based on the principle of minimum loss until the model converges; Step 5: Execute the generated maintenance action sequence and continuously optimize the decision-maker based on the equipment status after execution.

2. The intelligent digital management method for equipment assets according to claim 1, characterized in that, The process of transforming multimodal time-series data streams into high-dimensional feature vectors involves the following sub-steps: Perform time alignment and preprocessing on the data stream acquired by the sensor cluster; The preprocessed data stream is input into a pre-trained dedicated feature extraction neural network to extract the vibration feature vector, temperature feature vector and acoustic feature vector of the equipment; The extracted modal feature vectors are concatenated to generate a high-dimensional feature vector.

3. The intelligent digital management method for equipment assets according to claim 1, characterized in that, The decision-maker is continuously optimized based on the device status after execution, which is specifically divided into the following sub-steps: The current health status of the device is calculated based on the multimodal time-series data stream fed back by the sensor cluster; If the health status exceeds the predicted health status, then the decision weight of each type of maintenance action in the current maintenance action sequence will be increased. If the health status does not reach the predicted health status, the prediction bias of the evaluation network is corrected.

4. An intelligent digital management system for equipment assets, characterized in that, The method for implementing the intelligent digital management of equipment assets as described in any one of claims 1-3 includes: a data flow conversion module, a health status vector generation module, a health degradation trajectory generation module, a maintenance action sequence generation module, and a maintenance action sequence execution module. The data stream conversion module is used to collect multimodal time-series data streams emitted by physical devices using a sensor cluster and convert them into high-dimensional feature vectors. The health status vector generation module is used to deduce the internal state variables of the device at the current moment using high-dimensional feature vectors, and then combine them to generate the device health status vector. The health degradation trajectory generation module is used to analyze the device health status vector under time sequence arrangement and generate health degradation trajectory; The maintenance action sequence generation module is used to input the health degradation trajectory into the decision-maker and generate the optimal maintenance action sequence for the equipment. The maintenance action sequence execution module is used to execute the generated maintenance action sequence and continuously optimize the decision-maker based on the equipment status after execution.

5. The intelligent digital management system for equipment assets according to claim 4, characterized in that, The data stream transformation module specifically includes: a data stream preprocessing submodule, a feature vector extraction submodule, and a high-dimensional feature vector generation submodule; The data stream preprocessing submodule is used to perform time alignment and preprocessing on the data streams acquired by the sensor cluster. The feature vector extraction submodule is used to input the preprocessed data stream into a pre-trained dedicated feature extraction neural network to extract the vibration feature vector, temperature feature vector and acoustic feature vector of the equipment; The high-dimensional feature vector generation submodule is used to concatenate the extracted modal feature vectors to generate a high-dimensional feature vector.

6. The intelligent digital management system for equipment assets according to claim 4, characterized in that, The health degradation trajectory generation module specifically includes: a feature enhancement submodule, a health sequence generation submodule, and a health degradation trajectory improvement submodule; The feature enhancement submodule is used to adaptively enhance the device health status vector based on contextual data of the device's operating environment. The health sequence generation submodule is used to calculate the health of the device at each time step t based on the degradation factors of each dimension of the device health status vector, and organize it into the original health sequence. The Health Degradation Trajectory Improvement Submodule is used to input the original health sequence into the GPR model and output the complete health degradation trajectory.

Images