A remote operation and maintenance and fault prediction system of an intelligent internet of things device
By combining multi-dimensional data collection and digital twin models with an improved time series-support vector machine fusion algorithm and reinforcement learning-genetic algorithm, the problem of insufficient linkage between the IoT device status prediction module and the maintenance decision module is solved, realizing precise and dynamic maintenance management and improving prediction accuracy and the scientific nature of maintenance plans.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- LEGER TECH SERVICES LTD
- Filing Date
- 2025-10-15
- Publication Date
- 2026-06-19
AI Technical Summary
In existing technologies, the status prediction module and maintenance decision module of IoT devices lack sufficient linkage. The prediction results are only used as reference information for decision-making rather than core driving parameters, resulting in a deviation between maintenance decisions and the actual status of the equipment, making it difficult to achieve accurate and dynamic maintenance management.
By employing multi-dimensional data acquisition, digital twin model construction, an improved time series-support vector machine fusion algorithm, and a reinforcement learning-genetic algorithm combination framework, combined with production plans and real-time data, the prediction parameter weights and constraints are dynamically adjusted to achieve deep linkage between state prediction and maintenance decisions.
It improves the prediction accuracy of IoT device operating parameter change trends, fault types, and remaining service life, optimizes maintenance time, methods, and personnel plans, reduces maintenance costs and production losses, and ensures accurate matching of maintenance decisions with equipment status.
Smart Images
Figure CN122242990A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of Internet of Things (IoT) devices, specifically to a remote operation and maintenance and fault prediction system for intelligent IoT devices. Background Technology
[0002] IoT devices are critical equipment for data processing in industrial production, and their operating status directly affects the continuity and safety of the production system. Monitoring operating parameters such as vibration, temperature, and pressure, combined with predictive algorithms to assess equipment failure risks and remaining lifespan, and then developing maintenance strategies accordingly, is the core technological approach to ensure their stable operation.
[0003] In existing technologies, sensors are typically used to collect operational data from IoT devices. Time series analysis or machine learning algorithms are then used to predict the status of the devices. This data is then combined with production plans to develop maintenance solutions manually or semi-automatically. Some technologies introduce digital twin models to assist in simulating the operational status of the devices, thereby enhancing the reference value of prediction and decision-making.
[0004] The most critical shortcoming of the existing technology is that the linkage between the status prediction module and the maintenance decision module is insufficient. The prediction results are only used as reference information for decision-making rather than core driving parameters. The constraints are not dynamically adjusted in real time to keep up with the actual status of the equipment, resulting in a deviation between the maintenance decision and the actual operating status of the equipment. It is difficult to achieve accurate and dynamic maintenance management. In view of this, we propose a remote operation and maintenance and fault prediction system for intelligent IoT devices. Summary of the Invention
[0005] To address the shortcomings of existing technologies, this invention provides a remote operation and maintenance and fault prediction system for intelligent IoT devices, which solves the problem that the linkage between the state prediction module and the maintenance decision module is insufficient in existing technologies, and the prediction results are only used as reference information for decision-making rather than core driving parameters.
[0006] To achieve the above objectives, the present invention provides a remote operation and maintenance and fault prediction system for intelligent Internet of Things (IoT) devices, comprising the following steps:
[0007] The multi-dimensional data acquisition module is used to collect vibration, temperature, pressure, flow rate, speed signals, and ambient temperature, humidity, and dust concentration data of IoT devices during operation. It also collects images of the IoT device's exterior and internal flow field.
[0008] The digital twin model building module, based on the initial data from IoT device design drawings, material properties, manufacturing process information, and multi-dimensional data acquisition modules, constructs a digital twin model that includes a geometric model, a physical model, a behavioral model, and a rule model.
[0009] The status prediction module utilizes real-time data from a digital twin model and a multi-dimensional data acquisition module, and employs an improved time series-support vector machine fusion algorithm to predict future operating parameter trends, fault types, fault occurrence times, and remaining lifespan of IoT devices.
[0010] The intelligent maintenance decision-making module, based on the prediction results of the status prediction module and combined with production plans and maintenance costs, adopts a reinforcement learning-genetic algorithm combined framework to formulate maintenance time, methods and personnel plans.
[0011] The data storage and management module stores the raw data from the multi-dimensional data acquisition module, the digital twin model data, the results from the status prediction module, and the intelligent maintenance decision-making module solution, and provides data query, statistics, and analysis functions.
[0012] The human-computer interaction module displays the operating status, prediction results, and maintenance decision information of IoT devices, and receives system parameters and production plan information set by the user.
[0013] Preferably, the vibration sensor in the multi-dimensional data acquisition module is a three-dimensional accelerometer that simultaneously acquires vibration data in three directions; the temperature sensor is a distributed fiber optic temperature sensor that continuously monitors the temperature of key parts of the IoT device; and the image acquisition device is a high-definition industrial camera with infrared imaging function.
[0014] Preferably, the geometric model in the digital twin model construction module restores the three-dimensional structure of the IoT device at a 1:1 scale, the physical model includes multi-physics coupling analysis of flow field, temperature field, and stress field, the behavioral model simulates the operating behavior under different working conditions, and the rule model covers fault diagnosis rules and maintenance strategy rules.
[0015] Preferably, the digital twin model building module incorporates machine learning algorithms during the model building process to optimize model parameters by learning from historical data.
[0016] Preferably, the improved time series-support vector machine fusion algorithm of the state prediction module includes a dynamic weight allocation unit to perform real-time weighted fusion of time series analysis results and neural network output values.
[0017] Preferably, the fusion algorithm of the state prediction module embeds an adaptive correction factor for operating conditions, which dynamically adjusts the weights of the prediction parameters according to the real-time load changes of the IoT devices.
[0018] Preferably, in the reinforcement learning-genetic algorithm combined framework of the intelligent maintenance decision module, the reinforcement learning unit generates an initial maintenance strategy set, and the genetic algorithm iteratively optimizes the set.
[0019] Preferably, the combined framework of the intelligent maintenance decision module includes a constraint dynamic update unit, which incorporates production plan adjustment information in real time to correct decision parameters.
[0020] Preferably, the data storage and management module adopts distributed database technology to perform regular backup and recovery operations on the stored data.
[0021] Preferably, the human-computer interaction module has a remote control function, allowing users to send control commands through the interface to adjust system operating parameters.
[0022] This invention provides a remote operation and maintenance and fault prediction system for intelligent Internet of Things (IoT) devices. It has the following beneficial effects:
[0023] 1. This invention uses an improved time series-support vector machine fusion algorithm, combining ARIMA's ability to capture time series trends with SVM's ability to fit nonlinear patterns, and introduces dynamic weight allocation and operating condition adaptive correction factors to dynamically adjust the weights of prediction parameters. This improves the accuracy of predicting the changing trends of operating parameters, fault types, and remaining service life of IoT devices, providing a reliable basis for maintenance decisions.
[0024] 2. This invention utilizes a reinforcement learning-genetic algorithm combined framework. Reinforcement learning generates the initial maintenance strategy, and the genetic algorithm iteratively optimizes it. Combined with the RUL and failure probability of the state prediction module, the constraints are dynamically updated to ensure that maintenance decisions are based on the future state of the equipment. This achieves optimization of maintenance time, methods, and personnel plans, reducing maintenance costs and production losses.
[0025] 3. This invention strengthens the deep linkage between the state prediction module and the intelligent maintenance decision module, uses the RUL of the state prediction and the failure probability as inputs to the reinforcement learning state space, and adjusts the state transition probability with the working condition adaptive correction factor. This makes the reliability index in the fitness function of the genetic algorithm rely on the RUL for calculation, and ensures that the maintenance decision does not deviate from the actual state of the equipment through constraints. This achieves accurate matching between prediction and decision, and greatly improves the scientificity and feasibility of the maintenance plan. Attached Figure Description
[0026] Figure 1 This is a flowchart of a digital twin-based IoT device status prediction and maintenance system. Detailed Implementation
[0027] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0028] Example:
[0029] Please see the appendix Figure 1 This invention provides a remote operation and maintenance and fault prediction system for intelligent Internet of Things (IoT) devices, comprising the following steps:
[0030] The multi-dimensional data acquisition module is used to collect vibration, temperature, pressure, flow rate, and rotation speed signals, as well as ambient temperature, humidity, and dust concentration data during the operation of IoT devices. It also collects external images and internal flow field images of IoT devices. The vibration sensor in the multi-dimensional data acquisition module is a three-dimensional accelerometer sensor that simultaneously collects vibration data in three directions; the temperature sensor is a distributed fiber optic temperature sensor that continuously monitors the temperature of key parts of the IoT device; and the image acquisition device is a high-definition industrial camera with infrared imaging capabilities.
[0031] The digital twin model building module, based on the initial data from IoT device design drawings, material properties, manufacturing process information, and multi-dimensional data acquisition modules, constructs a digital twin model that includes a geometric model, a physical model, a behavioral model, and a rule model.
[0032] The geometric model in the digital twin model building module restores the three-dimensional structure of the IoT device at a 1:1 scale. The physical model includes multi-physics coupling analysis of flow field, temperature field, and stress field. The behavioral model simulates the operating behavior under different working conditions. The rule model covers fault diagnosis rules and maintenance strategy rules. The digital twin model building module introduces machine learning algorithms during the model building process to optimize model parameters by learning from historical data.
[0033] The state prediction module utilizes a digital twin model and real-time data from a multi-dimensional data acquisition module. It employs an improved time-series-support vector machine fusion algorithm to predict future operating parameter trends, fault types, fault occurrence times, and remaining lifespan of IoT devices. The improved time-series-support vector machine fusion algorithm includes a dynamic weight allocation unit that performs real-time weighted fusion of time-series analysis results and neural network output values. The fusion algorithm of the state prediction module embeds an adaptive correction factor based on operating conditions, dynamically adjusting the weights of prediction parameters according to real-time load changes of the IoT device. This includes the following algorithms:
[0034] ARIMA model construction
[0035] For a single-parameter time series yt (such as vibration amplitude), an ARIMA(p,d,q) model is established, and non-stationarity is eliminated through differencing:
[0036]
[0037] in:
[0038] B is the lag operator. This represents a p-order autoregressive coefficient polynomial;
[0039] Represents a q-th order moving average coefficient polynomial;
[0040] d is the difference order, which makes the sequence stationary; is the white noise error term, and c is a constant term. The optimal orders p, d, and q are determined using the AIC / BIC criterion to achieve short-term prediction of parameter variation trends.
[0041] SVM regression model construction
[0042] The prediction function uses a radial basis kernel function:
[0043]
[0044]
[0045] in:
[0046] The Lagrange multipliers are determined by solving a quadratic programming problem;
[0047] These are kernel parameters that control the width of the kernel function;
[0048] b is the bias term.
[0049] Dynamic weight allocation mechanism
[0050] The weights are dynamically adjusted based on the historical performance of the two models:
[0051]
[0052] in:
[0053] For the ARIMA model in the past The mean square error at each time point.
[0054] To adjust the parameter, a larger value indicates greater sensitivity to error;
[0055] When ARIMA predictions are more accurate The SVM weights approach 1; conversely, the weights increase to achieve adaptive fusion.
[0056] Fusion prediction implementation
[0057] Combine the prediction results of the two models according to their weights:
[0058]
[0059] This mechanism combines ARIMA's ability to capture time-series trends with SVM's ability to fit nonlinear patterns, thereby improving prediction accuracy.
[0060] Operating condition adaptive correction
[0061] Introducing a condition-based correction factor CFi(t), the prediction results are dynamically adjusted based on real-time load:
[0062]
[0063] in:
[0064] For example, in the flow correction factor, For current traffic, This is the rated flow rate.
[0065] The weighting coefficients are obtained through training with historical data. For example, the correction weights for flow rate and pressure are 0.3 and 0.7, respectively. This mechanism enables the system to maintain high-precision predictions even under varying operating conditions.
[0066] Fault type prediction
[0067] Based on the multi-class SVM model, the probability distribution of each type of fault is output:
[0068]
[0069] in:
[0070] For the first The decision function for class-specific faults converts the output into probabilities through the Softmax function, and takes the category corresponding to the maximum value as the prediction result to achieve the classification and identification of faults such as bearing wear and impeller cavitation.
[0071] The intelligent maintenance decision-making module, based on the prediction results of the state prediction module and combined with production plans and maintenance costs, employs a reinforcement learning-genetic algorithm combined framework to formulate maintenance time, methods, and personnel plans. In this framework, the reinforcement learning unit generates an initial set of maintenance strategies, and the genetic algorithm iteratively optimizes the set. The combined framework also includes a constraint dynamic update unit that incorporates production plan adjustment information in real time to correct decision parameters, including the following algorithms:
[0072] 1. Reinforcement Learning-Genetic Algorithm Combinatorial Framework
[0073] 1.1 Reinforcement Learning Unit (Generating Initial Policy Set) Markov Decision Process (MDP) Model:
[0074]
[0075] in:
[0076] S: State space, which integrates the output of the state prediction module (such as remaining useful life RUL, probability of failure) and production planning information (such as allowable downtime window), and is the basic input for decision-making;
[0077] A: Action space, which includes specific maintenance decision options, such as maintenance method (preventive maintenance can prevent the fault from escalating, and corrective maintenance is used for post-fault repair), maintenance time point, and personnel configuration for maintenance;
[0078] P: Simulate the impact of different maintenance actions on equipment status using digital twin models (e.g., how the probability of equipment failure changes after preventive maintenance is performed).
[0079] R: Measures the quality of maintenance actions, taking into account overall maintenance costs (such as spare parts costs and labor costs), production losses (output reduction due to downtime), and the benefits of extended equipment lifespan.
[0080] Learning Algorithms:
[0081]
[0082] α: Learning rate, which controls the weight of new experiences;
[0083] Values range from 0 to 1, reflecting the degree of importance attached to future rewards, for example... 0.9 indicates that the reward for the 10th step in the future is only equivalent to the current reward. times;
[0084] The part inside the brackets is "temporal difference error", which reflects the gap between the current expected reward and the actual reward plus the future optimal reward. The strategy is optimized by iteratively reducing the error.
[0085] Initial policy set generation:
[0086]
[0087] That is, select the optimal action for each state from the Q value to form multiple initial maintenance strategies (such as maintenance schemes at different time points).
[0088] 1.2 Genetic Algorithm Unit (Iterative Optimization Strategy)
[0089] Chromosome coding:
[0090] Chromosome Methods, personnel ,personnel This transforms maintenance strategies into computable "gene sequences."
[0091] : The specific time point for maintenance execution;
[0092] Methods: such as "preventive maintenance" and "replacing critical components";
[0093] Personnel ID: The personnel ID of those involved in maintenance, ensuring manpower matching.
[0094] Fitness function:
[0095]
[0096] Total maintenance costs, including manpower, spare parts, and production losses;
[0097] Post-maintenance equipment reliability is calculated by the RUL of the condition prediction module;
[0098] The adaptability of the strategy to changes in the production plan;
[0099] , , Weighting, adjusted according to actual needs (e.g., increased during peak production seasons). Prioritize flexibility).
[0100] Select operation:
[0101] Representation Strategy The probability of being selected for the next generation is that strategies with higher fitness are more likely to be retained, similar to "survival of the fittest".
[0102] Cross operation: Simulates gene recombination, combining parts of two parent strategies to generate offspring strategies;
[0103] Crossover probability
[0104] Mutation operation: Randomly modify some parameters in the strategy;
[0105] Mutation probability helps avoid the strategy getting trapped in local optima and ensures exploratory nature.
[0106] Constraint dynamic update unit
[0107] Production planning constraints:
[0108] The formula is Constraint1: Maintenance time must be limited to within the window allowed by the production plan.
[0109] Updated in real time according to the production plan (e.g., when urgent orders increase). (Possibly earlier).
[0110] Resource constraints:
[0111] The formula is Constraint2: Ensure that the resources required for maintenance do not exceed the available amount.
[0112] : Strategy For the Requirements for similar resources (e.g., "2 technicians" or "1 pump shaft spare part");
[0113] The total amount of resources currently available for allocation.
[0114] State prediction linkage constraints:
[0115] The formula is Constraint3: Core linkage constraints ensure that maintenance is completed before equipment failure.
[0116] The remaining lifespan output by the status prediction module (e.g., predicting that the equipment can still operate normally for 5 days).
[0117] The interval between the current time and the planned maintenance (e.g., if maintenance is planned in 3 days, the interval is 3 days) must meet the condition that the interval is ≤ RUL to avoid the equipment failure decision parameter correction mechanism before maintenance.
[0118] Decision parameter correction mechanism
[0119] Dynamic weight adjustment: The formula is as follows The weights in the fitness function are adjusted based on real-time feedback.
[0120] (Adjusting step size): Controls the magnitude of weight changes;
[0121] Fitness versus weights The partial derivative reflects the impact of weight changes on strategy evaluation (e.g., when maintenance costs exceed the budget, it will increase). (With a greater emphasis on cost).
[0122] Integration of state prediction results: The formula is as follows The production loss is linked to the failure probability of the status prediction module.
[0123] Production loss = (Downtime, probability of failure) The higher the probability of failure (e.g., 80% probability of equipment failure during predicted maintenance), the greater the emergency repair costs and production losses caused by downtime.
[0124] Failure probability The failure probability at a specific point in time output by the state prediction module;
[0125] Summary of the connection with the algorithms mentioned above
[0126] The RUL and failure probability of the state prediction module are directly used as the state space for reinforcement learning. The input affects the learning of "state-action" value;
[0127] The adaptive correction factor adjusts the state transition probability. This makes reinforcement learning more closely aligned with the actual operating conditions of the equipment;
[0128] In the fitness function of the genetic algorithm, the reliability index depends entirely on the RUL calculation of the state prediction module;
[0129] Constraint is the core of connecting the two modules, ensuring that maintenance decisions are based on predictions of the future state of the equipment and avoiding a disconnect between decisions and the actual state.
[0130] Final decision output
[0131] Optimal Strategy It is necessary to simultaneously meet the requirements of minimizing costs and production losses, while ensuring that maintenance time is within the RUL allowable range, and finally output specific maintenance time, methods and personnel plans;
[0132] The data storage and management module stores the raw data from the multi-dimensional data acquisition module, the digital twin model data, the results from the status prediction module, and the intelligent maintenance decision-making module scheme. It provides data query, statistics, and analysis functions. The data storage and management module adopts distributed database technology to perform regular backup and recovery operations on the stored data.
[0133] The human-machine interaction module displays the operating status, prediction results, and maintenance decision information of IoT devices, and receives system parameters and production plan information set by the user. The human-machine interaction module has remote control function, and the user can send control commands through the interface to adjust the system operating parameters.
[0134] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.
Claims
1. A remote operation and maintenance and fault prediction system for intelligent Internet of Things (IoT) devices, characterized in that, Includes the following steps: The multi-dimensional data acquisition module is used to collect vibration, temperature, pressure, flow rate, speed signals, and ambient temperature, humidity, and dust concentration data of IoT devices during operation. It also collects images of the IoT device's exterior and internal flow field. The digital twin model building module, based on the initial data from IoT device design drawings, material properties, manufacturing process information, and multi-dimensional data acquisition modules, constructs a digital twin model that includes a geometric model, a physical model, a behavioral model, and a rule model. The status prediction module utilizes real-time data from a digital twin model and a multi-dimensional data acquisition module, and employs an improved time series-support vector machine fusion algorithm to predict future operating parameter trends, fault types, fault occurrence times, and remaining lifespan of IoT devices. The intelligent maintenance decision-making module, based on the prediction results of the status prediction module and combined with production plans and maintenance costs, adopts a reinforcement learning-genetic algorithm combined framework to formulate maintenance time, methods and personnel plans. The data storage and management module stores the raw data from the multi-dimensional data acquisition module, the digital twin model data, the results from the status prediction module, and the intelligent maintenance decision-making module solution, and provides data query, statistics, and analysis functions. The human-computer interaction module displays the operating status, prediction results, and maintenance decision information of IoT devices, and receives system parameters and production plan information set by the user.
2. The remote operation and maintenance and fault prediction system for intelligent IoT devices according to claim 1, characterized in that, The vibration sensor in the multi-dimensional data acquisition module is a three-dimensional accelerometer that simultaneously acquires vibration data in three directions; the temperature sensor is a distributed fiber optic temperature sensor that continuously monitors the temperature of key parts of IoT devices; and the image acquisition device is a high-definition industrial camera with infrared imaging capabilities.
3. The remote operation and maintenance and fault prediction system for intelligent IoT devices according to claim 1, characterized in that, The geometric model in the digital twin model construction module restores the three-dimensional structure of the IoT device at a 1:1 scale. The physical model includes multi-physics coupling analysis of flow field, temperature field, and stress field. The behavioral model simulates the operating behavior under different working conditions. The rule model covers fault diagnosis rules and maintenance strategy rules.
4. The remote operation and maintenance and fault prediction system for intelligent IoT devices according to claim 1, characterized in that, The digital twin model building module incorporates machine learning algorithms during the model building process to optimize model parameters by learning from historical data.
5. The remote operation and maintenance and fault prediction system for intelligent IoT devices according to claim 1, characterized in that, The improved time series-support vector machine fusion algorithm of the state prediction module includes a dynamic weight allocation unit that performs real-time weighted fusion of time series analysis results and neural network output values.
6. The remote operation and maintenance and fault prediction system for intelligent IoT devices according to claim 1, characterized in that, The fusion algorithm of the state prediction module embeds an adaptive correction factor for operating conditions, which dynamically adjusts the weights of prediction parameters according to the real-time load changes of IoT devices.
7. The remote operation and maintenance and fault prediction system for intelligent IoT devices according to claim 1, characterized in that, In the reinforcement learning-genetic algorithm combined framework of the intelligent maintenance decision module, the reinforcement learning unit generates an initial set of maintenance strategies, and the genetic algorithm iteratively optimizes the set.
8. The remote operation and maintenance and fault prediction system for intelligent IoT devices according to claim 1, characterized in that, The combined framework of the intelligent maintenance decision module includes a constraint dynamic update unit that incorporates production plan adjustment information in real time to correct decision parameters.
9. The remote operation and maintenance and fault prediction system for intelligent IoT devices according to claim 1, characterized in that, The data storage and management module uses distributed database technology to perform regular backup and recovery operations on the stored data.
10. A remote operation and maintenance and fault prediction system for intelligent Internet of Things (IoT) devices according to claim 1, characterized in that, The human-computer interaction module has remote control capabilities, allowing users to send control commands through the interface to adjust system operating parameters.