Intelligent maintenance and inspection method and device for equipment and equipment
By optimizing equipment fault diagnosis using digital twin models and fuzzy quantum deep learning algorithms, the problems of low accuracy in equipment fault diagnosis and low maintenance efficiency have been solved. Real-time monitoring of equipment status and automatic fault identification have been achieved, reducing maintenance costs and ensuring production safety.
Patent Information
- Application Number
- CN202511595193.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-03
- Publication Date
- 2026-02-10
AI Technical Summary
Existing technologies suffer from low accuracy in equipment fault diagnosis, model training is prone to getting stuck in local optima, low maintenance and inspection efficiency, difficulty in ensuring data quality, and imbalanced training datasets.
A digital twin model combined with fuzzy quantum deep learning algorithm is adopted. High-quality operational data is obtained by constructing a digital twin model of the device, and the model parameters are optimized by using fuzzy quantum deep learning algorithm. Fault diagnosis is performed by combining CNN-LSTM model, and fuzzy search, balanced spiral and greedy quantum search strategies are adopted to improve model training efficiency.
It enables real-time monitoring of equipment status and automatic identification of fault types, generates intelligent maintenance reports, reduces unplanned downtime, lowers maintenance costs, and ensures production safety.
Smart Images
Figure CN121504424A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of industrial equipment maintenance technology, and more specifically, to an intelligent equipment maintenance and inspection method, device, and equipment. Background Technology
[0002] In modern industrial production, the stable operation of large-scale critical equipment (such as wind turbines, CNC machine tools, and chemical reactors) is crucial for ensuring production efficiency and product quality. Traditional equipment maintenance methods mainly rely on periodic inspections and reactive repairs, which have many drawbacks. Periodic inspections may lead to "over-maintenance," increasing unnecessary costs and failing to prevent sudden failures; reactive repairs often result in equipment downtime, production interruptions, and significant economic losses, and may even trigger safety accidents.
[0003] With the development of sensor technology and the Internet of Things (IoT), predictive maintenance based on condition monitoring has become a research hotspot. By deploying sensors on equipment to collect operational data such as vibration, temperature, and current, and then analyzing this data using machine learning or deep learning algorithms, early warning and diagnosis of faults can be achieved. However, existing technologies still face the following challenges: Data quality and acquisition challenges: Directly collecting high-frequency, multi-dimensional data on physical devices can be affected by harsh operating conditions, sensor malfunctions, and other factors, making it difficult to guarantee data quality. At the same time, real-world fault samples are scarce, and annotation costs are high, leading to an imbalanced training dataset.
[0004] Model performance bottlenecks: Traditional deep learning models (such as CNN (Convolutional Neural Network) and LSTM (Long Short-Term Memory)) are prone to getting stuck in local optima when processing complex, nonlinear equipment fault data, resulting in slow convergence speed and weak generalization ability. Their ability to extract features from early, subtle faults is limited, and diagnostic accuracy needs improvement.
[0005] Low algorithm optimization efficiency: Algorithms used to optimize deep learning model parameters (such as gradient descent) are sensitive to initial values and have a single search strategy, making it difficult to efficiently find the global optimum in a huge parameter space. Summary of the Invention
[0006] This application aims to provide an intelligent equipment maintenance and inspection method, apparatus, and equipment, which are intended to solve the problems of low accuracy in equipment fault diagnosis, easy trapping of model training in local optima, and low maintenance and inspection efficiency in the prior art.
[0007] The first aspect of this application provides an intelligent equipment maintenance and inspection method, including: Construct a digital twin model corresponding to the target device, and detect the corresponding target operation data through the digital twin model of the target device; Obtain the actual fault labels corresponding to the target operating data input by the staff, and construct prior experience data by combining the target operating data and the corresponding actual fault labels; The fuzzy quantum deep learning algorithm is used to learn the data relationships in the prior experience data to obtain an intelligent maintenance and inspection model for the equipment; The real-time operating data of the target device is detected by the digital twin model corresponding to the target device, and the intelligent maintenance and inspection model of the device is called to identify the real-time operating data to determine the intelligent maintenance and inspection identification result of the device. Based on the intelligent maintenance and inspection results, an intelligent maintenance and inspection report is generated for the equipment, and the report is pushed to the equipment designated by the staff.
[0008] In one possible implementation, the target operating data includes at least vibration signals, temperature signals, current signals, and voltage signals of the target device at multiple consecutive sampling time points.
[0009] In one possible implementation, a fuzzy quantum deep learning algorithm is used to learn the data relationships in the prior experience data to obtain an intelligent maintenance and inspection model for the equipment, including: A target deep learning model is constructed using a CNN-LSTM model, and the model parameters of the target deep learning model are initialized to obtain the training population; For any individual in the training population, the fitness of the individual is obtained based on the prior experience data, and the optimal individual is determined based on the fitness. For any given individual, based on the optimal individual, and using a fuzzy search strategy, a fuzzy search is performed on the individual's solution space to obtain the first target individual; For any first target individual, a balanced spiral search strategy is used to perform a local augmentation search in the solution space of the first target individual to obtain the second target individual; For any number of second target individuals, a greedy quantum search strategy is used to perform a global augmentation search of the solution space of the second target individuals to obtain the third target individuals; Obtain the current number of training iterations, and obtain the training status based on the current number of training iterations; wherein, the training status includes a training completed status or a training incomplete status; When the training state is the training completed state, the final parameters of the target deep learning model are determined according to the third target individual to obtain the intelligent maintenance and inspection model of the equipment. When the training state is incomplete, the process returns to the step of determining the optimal individual based on the third target individual, thus entering the next training process.
[0010] In one possible implementation, initializing the model parameters of the target deep learning model to obtain a training population includes: initializing the model parameters of the target deep learning model using a random initialization strategy or a chaotic mapping initialization strategy to obtain a training population containing multiple individuals; wherein, the individuals include some or all of the model parameters to be trained of the target deep learning model.
[0011] In one possible implementation, obtaining the fitness of an individual based on the prior experience data, and determining the optimal individual based on the fitness, includes: For any individual, the model parameters contained in the individual are applied to the target deep learning model to obtain the target deep learning model after applying the parameters; The target running data in the prior experience data is used as the actual input of the target deep learning model after applying the parameters, and the real fault labels in the prior experience data are used as the expected output of the target deep learning model after applying the parameters to obtain the cross-entropy loss function value. The cross-entropy loss function value is added to a preset constant term to obtain a non-zero data term, and the reciprocal of the non-zero data term is taken to obtain the fitness of the individual. Obtain the fitness of all individuals and determine the individual with the highest fitness as the optimal individual.
[0012] In one possible implementation, based on the optimal individual, and employing a fuzzy search strategy to perform a fuzzy search of the solution space to obtain the first target individual, the method includes: Based on the fitness of the individual, the first fuzzification factor and the second fuzzification factor are obtained as follows:
[0013]
[0014] in, Indicates the first i The first fuzziness factor for each individual Indicates the first i The second fuzzification factor for each individual, Indicates the first i The fitness of an individual i =1,2,…,MP, where MP represents the total number of individuals. This represents the fitness of the optimal individual. This represents the fitness of the worst individual, which is the individual with the lowest fitness. a This represents the constant term in [0.8, 1). Based on the first fuzzification factor and the second fuzzification factor, the fuzzification control coefficients corresponding to the training population are obtained as follows:
[0015] in, Represents the fuzzy control coefficient. express and The maximum value in; Based on the fuzzification control coefficients, the first search factor and the second search factor are obtained as follows:
[0016]
[0017] in, Indicates the first search factor. Indicates the second search factor. This represents the first random number, which is either 0 or 1. This represents a second random number that is either 0 or 1. Indicates the maximum number of training iterations; Based on the optimal individual, the first search factor, and the second search factor, a fuzzy search of the solution space is performed on the individual to obtain the first target individual:
[0018] in, Indicates the first t During the training process, the first i Individual, Indicates the first i The first target individual, Represents the optimal individual. Represents the natural constant. Indicates that among all individuals, except The first random individual outside of [the previous one].
[0019] In one possible implementation, a balanced spiral search strategy is used to perform a local enhancement search of the solution space on the first target individual to obtain the second target individual, including: The spiral search factor is obtained as follows:
[0020]
[0021] in, 'b' represents the spiral search factor, and 'b' represents the spiral path parameter. l The helix step size parameter is represented by cos, and cos represents the cosine function. This represents a third random number between (0,1). Represents the inverse hyperbolic tangent function; Based on the spiral search factor, a local enhancement search in the solution space is performed on the first target individual to obtain the second target individual:
[0022] in, Indicates the first t During the training process, the first j The first target individual, Indicates the first j A second target individual, Indicates the second search factor. This means that among all individuals in the first target category, excluding A second random individual other than; This represents the fourth random number, which is either 0 or 1.
[0023] In one possible implementation, a greedy quantum search strategy is used to perform a global augmentation search of the solution space on the second target individual to obtain the third target individual, including: The attenuation factor is obtained as follows:
[0024] in, The attenuation factor is represented by sine, and sin represents the sine function. Let represent the fitness of the best individual during the (t-1)th training iteration. Represents pi; Based on the attenuation factor, the global augmentation search of the solution space for the second target individual is as follows:
[0025] in, Indicates the first t During the training process, the first j A second target individual, Indicates the first j The second target individual after a global augmented search. Represents the optimal individual. This represents a random number between (-0.1, 0.1). This represents the mean individual obtained from all individuals with the second objective. Represents a random number between (0, 1); Determine whether the fitness of the second target individual after the global augmentation search has increased. If so, use the second target individual after the global augmentation search as the third target individual; otherwise, directly use the original second target individual as the third target individual.
[0026] A second aspect of this application provides an intelligent equipment maintenance and inspection device, comprising: The data acquisition module is used to construct a digital twin model corresponding to the target device and to detect the target operating data corresponding to the target device through the digital twin model corresponding to the target device. The data integration module is used to obtain the real fault labels corresponding to the target operation data input by the staff, and to jointly construct prior experience data by the target operation data and its corresponding real fault labels. The deep learning module is used to learn the data relationships in the prior experience data using the fuzzy quantum deep learning algorithm to obtain an intelligent maintenance and inspection model for the equipment. The intelligent maintenance module is used to detect the corresponding real-time operating data of the target device through the digital twin model, and call the intelligent maintenance model of the device to identify the real-time operating data to determine the intelligent maintenance identification result of the device. The report generation module is used to generate an intelligent equipment maintenance report based on the intelligent equipment maintenance identification results, and push the intelligent equipment maintenance report to the equipment designated by the staff.
[0027] The third aspect of this application is an intelligent maintenance and inspection device for equipment, including a processor and a memory; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the intelligent equipment maintenance and inspection method as described in the first aspect.
[0028] Beneficial effects: This application provides an intelligent equipment maintenance and inspection method, device, and equipment that deeply integrates digital twin technology with advanced deep learning algorithms. The digital twin model provides a safe, controllable, and high-fidelity virtual environment, which can not only acquire high-quality operational data but also simulate various extreme working conditions, providing rich data support for model training. During the training process, fuzzy quantum deep learning algorithms are used to learn the data relationships in prior experience data, improving the model's recognition accuracy. The trained intelligent maintenance and inspection model can monitor equipment status in real time, automatically identify fault types and severity, and automatically generate reports to push to relevant personnel, realizing proactive early warning. This makes equipment maintenance and inspection work more intelligent, efficient, and forward-looking, minimizing unplanned downtime, reducing maintenance costs, and ensuring production safety. Attached Figure Description
[0029] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0030] Figure 1 This is a flowchart of an intelligent equipment maintenance method proposed in one embodiment of this application; Figure 2 This is a schematic diagram of the structure of an intelligent equipment maintenance and inspection device according to an embodiment of this application; Figure 3 This is a schematic diagram of the structure of an intelligent equipment maintenance and inspection device according to an embodiment of this application; Explanation of reference numerals in the attached diagram: 201-Data acquisition module, 202-Data integration module, 203-Deep learning module, 204-Intelligent maintenance module, 205-Report generation module, 301-Memory, 302-Processor, 303-Bus. Detailed Implementation
[0031] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.
[0032] like Figure 1 As shown in the figure, this application provides an intelligent equipment maintenance method, including: S101. Construct a digital twin model corresponding to the target device, and detect the target operation data corresponding to the target device through the digital twin model corresponding to the target device.
[0033] In this embodiment, the target equipment can be key equipment such as CNC machine tools on an industrial production line, gearboxes of wind turbine generators, or pumps in a chemical plant. First, a digital twin model highly consistent with the physical equipment in terms of geometry, physics, behavior, and rules is constructed using 3D modeling, physical simulation, and data-driven methods. This model can receive data from the physical equipment's sensors in real time and update its own status synchronously. Through this digital twin model, high-fidelity operating data of the equipment under various working conditions can be obtained safely and conveniently.
[0034] The target operating data includes at least vibration, temperature, current, and voltage signals of the target equipment at multiple consecutive sampling time points. This data comprehensively reflects the mechanical, electrical, and thermodynamic state of the equipment and serves as a crucial basis for fault diagnosis. For example, changes in the vibration signal spectrum can indicate mechanical faults such as bearing wear or rotor imbalance; abnormal temperatures may indicate poor lubrication or overload; and fluctuations in current and voltage can reflect problems with motor load and the electrical system.
[0035] S102. Obtain the real fault labels corresponding to the target operating data input by the staff, and construct prior experience data by combining the target operating data and the corresponding real fault labels.
[0036] The acquired target operational data needs to be labeled before it can be used for supervised learning. Staff (such as senior engineers) can analyze historical maintenance records, expert experience, or fault scenarios simulated by the digital twin model to label a segment of operational data with realistic fault labels, such as normal, bearing inner race fault, gear tooth breakage, or motor overheating. Collecting these labeled data pairs (target operational data, realistic fault labels) constitutes the prior experience dataset used for model training. This dataset forms the knowledge foundation for building the intelligent maintenance and inspection model.
[0037] S103. Use fuzzy quantum deep learning algorithm to learn the data relationships in the prior experience data to obtain the intelligent maintenance and inspection model of the equipment.
[0038] Traditional gradient descent methods converge slowly when dealing with high-dimensional, non-convex parameter spaces of deep learning models. The fuzzy quantum deep learning algorithm presented in this application is a swarm intelligence optimization algorithm that, through a parallelized population search mechanism, can explore multiple regions simultaneously. The design of fuzzy search and spiral search strategies gives the algorithm strong global search capabilities in the early stages of training, quickly locating high-quality solution regions; in the later stages of training, it focuses on fine-grained local search, accelerating convergence. This adaptive search balancing mechanism significantly shortens the time required for the model to reach stability and high accuracy, improving the deployment efficiency of the maintenance and inspection system.
[0039] S104. Detect the corresponding real-time operating data of the target device through the digital twin model, and call the device intelligent maintenance and inspection model to identify the real-time operating data to determine the device intelligent maintenance and inspection identification result.
[0040] The trained intelligent equipment maintenance and inspection model is deployed online. Real-time operational data of the target equipment is acquired through the digital twin model (or directly from the physical equipment's sensors), and this real-time data is input into the model. The model automatically performs forward calculations and outputs a recognition result, typically a probability distribution indicating the probability of the equipment being in various states (e.g., normal, fault A, fault B, etc.). The state with the highest probability is taken as the final intelligent equipment maintenance and inspection recognition result.
[0041] S105. Generate an intelligent maintenance report for the equipment based on the intelligent maintenance identification results, and push the intelligent maintenance report to the equipment designated by the staff.
[0042] A structured maintenance report is automatically generated based on the identification results. The report may include: device ID, diagnosis time, diagnosis results (fault type, confidence level), key anomalies, and recommended maintenance measures. This report is then sent to designated maintenance engineers or managers via internal communication systems, email, SMS, or app push notifications so they can take timely action.
[0043] This application provides an intelligent equipment maintenance and inspection method that deeply integrates digital twin technology with advanced deep learning algorithms. The digital twin model provides a safe, controllable, and high-fidelity virtual environment, which can not only acquire high-quality operational data but also simulate various extreme working conditions, providing rich data support for model training. During the training process, fuzzy quantum deep learning algorithms are used to learn the data relationships in prior experience data, improving the model's recognition accuracy. The trained intelligent maintenance and inspection model can monitor equipment status in real time, automatically identify fault types and severity, and automatically generate reports to push to relevant personnel, realizing proactive early warning. This makes equipment maintenance and inspection work more intelligent, efficient, and forward-looking, minimizing unplanned downtime, reducing maintenance costs, and ensuring production safety.
[0044] In one possible implementation, the target operating data includes at least vibration signals, temperature signals, current signals, and voltage signals of the target device at multiple consecutive sampling time points.
[0045] It is worth noting that the target operating data mentioned above is merely an example. Different operating data can be set according to different devices. For example, motor speed can be collected, and water pump pressure can be collected, thereby achieving more accurate fault identification.
[0046] In one possible implementation, a fuzzy quantum deep learning algorithm is used to learn the data relationships in the prior experience data to obtain an intelligent maintenance and inspection model for the equipment, including: A target deep learning model is constructed using a CNN-LSTM model, and the model parameters of the target deep learning model (such as the connection weights between neural network layers) are initialized to obtain the training population.
[0047] For any individual in the training population, the fitness of the individual is obtained based on the prior experience data, and the optimal individual is determined based on the fitness.
[0048] For any given individual, based on the optimal individual, and using a fuzzy search strategy, a fuzzy search of the solution space is performed to obtain the first target individual.
[0049] For any first target individual, a balanced spiral search strategy is used to perform a local augmentation search in the solution space to obtain the second target individual.
[0050] For any number of second target individuals, a greedy quantum search strategy is used to perform a global augmentation search of the solution space of the second target individuals to obtain the third target individuals.
[0051] Obtain the current training iteration count, and determine the training status based on the current training iteration count. The training status includes either a training completed state or a training incomplete state.
[0052] For example, if the current number of training iterations is greater than or equal to the maximum number of training iterations, the training can be considered complete; otherwise, the training is considered incomplete.
[0053] When the training state is complete, the final parameters of the target deep learning model are determined based on the third target individual, resulting in an intelligent equipment maintenance and inspection model. For example, an optimal individual can be re-determined based on the third target individual, and the parameters contained in the re-determined optimal individual can be used as the final parameters of the target deep learning model to obtain the intelligent equipment maintenance and inspection model.
[0054] When the training state is incomplete, the process returns to the step of determining the optimal individual based on the third target individual, thus entering the next training process.
[0055] This application employs an innovative fuzzy quantum deep learning algorithm to train the maintenance and inspection model. This algorithm, through a fuzzy search strategy, dynamically adjusts the search step size and direction based on individual fitness, enhancing its global exploration capability and effectively avoiding the problem of traditional optimization algorithms prematurely getting trapped in local optima. By using a balanced spiral search strategy, simulating biological spiral predation behavior, it refines the local exploration of the neighborhood of the current optimal solution, improving the model's ability to converge to a high-precision solution. Through a greedy quantum search strategy, it introduces probabilistic perturbations and population mean information from quantum thinking, guiding individuals to jump out of local optima and towards the global optimum, while employing a greedy strategy to ensure that each iteration evolves towards a better solution. The organic combination of these three strategies results in a final intelligent equipment maintenance and inspection model with stronger feature extraction and pattern recognition capabilities, accurately identifying early and minor equipment faults, and achieving a significantly improved diagnostic accuracy compared to traditional methods.
[0056] Because the algorithm can more effectively find globally or near-globally optimal model parameters, the trained model is not merely a simple memorization of training data (prior experience data), but rather learns deeper, more universal fault characteristic patterns behind the data. Therefore, this model maintains high diagnostic accuracy and demonstrates excellent generalization ability when faced with new, unseen real-time operational data. Furthermore, the randomness and quantum perturbations introduced into the algorithm enhance the model's robustness to noise and minor data fluctuations.
[0057] In one possible implementation, initializing the model parameters of the target deep learning model to obtain a training population includes: initializing the model parameters of the target deep learning model using a random initialization strategy or a chaotic mapping initialization strategy to obtain a training population containing multiple individuals; wherein, the individuals include some or all of the model parameters to be trained of the target deep learning model. Therefore, the algorithm provided in the embodiments of this application can be used to train some of the model parameters to be trained, or the algorithm provided in the embodiments of this application can be used to train all the model parameters to be trained.
[0058] For example, the model parameters of the target deep learning model can be randomly initialized between the upper and lower bounds, and the initialized model parameters can be encoded into vectors to obtain individuals. This process can be repeated multiple times to obtain the training population. Random initialization strategies or chaotic mapping initialization strategies are relatively conventional techniques and existing technologies can be directly adopted, so they will not be elaborated here.
[0059] In one possible implementation, obtaining the fitness of an individual based on the prior experience data, and determining the optimal individual based on the fitness, includes: For any individual, the model parameters contained in the individual are applied to the target deep learning model to obtain the target deep learning model after applying the parameters; The target running data in the prior experience data is used as the actual input of the target deep learning model after applying the parameters, and the real fault labels in the prior experience data are used as the expected output of the target deep learning model after applying the parameters to obtain the cross-entropy loss function value. The cross-entropy loss function value is added to a preset constant (such as 0.01, 0.1 or 1) to obtain a non-zero data item, and the reciprocal of the non-zero data item is taken to obtain the fitness of the individual. Obtain the fitness of all individuals and determine the individual with the highest fitness as the optimal individual.
[0060] In one possible implementation, based on the optimal individual, and employing a fuzzy search strategy to perform a fuzzy search of the solution space to obtain the first target individual, the method includes: Based on the fitness of the individual, the first fuzzification factor and the second fuzzification factor are obtained as follows:
[0061]
[0062] in, Indicates the first i The first fuzziness factor for each individual Indicates the first i The second fuzzification factor for each individual, Indicates the first i The fitness of an individual i =1,2,…,MP, where MP represents the total number of individuals. This represents the fitness of the optimal individual. This represents the fitness of the worst individual, which is the individual with the lowest fitness. a This represents the constant term in [0.8, 1). Based on the first fuzzification factor and the second fuzzification factor, the fuzzification control coefficients corresponding to the training population are obtained as follows:
[0063] in, Represents the fuzzy control coefficient. express and The maximum value in; Based on the fuzzification control coefficients, the first search factor and the second search factor are obtained as follows:
[0064]
[0065] in, Indicates the first search factor. Indicates the second search factor. This represents the first random number, which is either 0 or 1. This represents a second random number that is either 0 or 1. Indicates the maximum number of training iterations; Based on the optimal individual, the first search factor, and the second search factor, a fuzzy search of the solution space is performed on the individual to obtain the first target individual:
[0066] in, Indicates the first t During the training process, the first i Individual, Indicates the first i The first target individual, Represents the optimal individual. Represents the natural constant. Indicates that among all individuals, except The first random individual outside of [the previous one].
[0067] This fuzzy search strategy introduces a fuzzification factor dynamically correlated with individual fitness, enabling adaptive adjustment of the search step size and effectively preventing the algorithm from prematurely getting trapped in local optima. Simultaneously, by dynamically adjusting the search factor through fuzzification control coefficients and combining update formulas for learning from the best individual and random exploration, the algorithm's development and exploration capabilities are balanced. Ultimately, this strategy guides the population to locate the region of the global optimum more quickly and accurately, laying a solid foundation for subsequent fine-tuning and thus improving the performance and training speed of the final model.
[0068] In one possible implementation, a balanced spiral search strategy is used to perform a local enhancement search of the solution space on the first target individual to obtain the second target individual, including: The spiral search factor is obtained as follows:
[0069]
[0070] in, represents the spiral search factor, and b represents the spiral path parameter, which can be set to 1 or a number that decreases uniformly from 2 to 0.1 as the number of training iterations changes. l The helix step size parameter is represented by cos, and cos represents the cosine function. This represents a third random number between (0,1). Represents the inverse hyperbolic tangent function; Based on the spiral search factor, a local enhancement search in the solution space is performed on the first target individual to obtain the second target individual:
[0071] in, Indicates the first t During the training process, the first j The first target individual, Indicates the first j A second target individual, Indicates the second search factor. This means that among all individuals in the first target category, excluding A second random individual other than; This represents the fourth random number, which is either 0 or 1.
[0072] This balanced spiral search strategy mimics the spiral approach behavior of organisms in nature, designing a unique search path. This path performs a fine-grained search around the current optimal solution, exploring multiple concentric layers from the outside in, ensuring comprehensive coverage of the optimal solution's neighborhood and effectively avoiding missing high-quality solutions. Simultaneously, the spiral search factor can be dynamically adjusted during training, achieving adaptive shrinkage of the search range. This allows the algorithm to explore a wide range in the early stages of training and focus on a minimal region in the later stages, accelerating convergence. This strategy systematically improves the accuracy of solutions, providing a powerful local augmentation method for finding the global optimum.
[0073] In one possible implementation, a greedy quantum search strategy is used to perform a global augmentation search of the solution space on the second target individual to obtain the third target individual, including: The attenuation factor is obtained as follows:
[0074] in, The attenuation factor is represented by sine, and sin represents the sine function. Let represent the fitness of the best individual during the (t-1)th training iteration. Represents pi; Based on the attenuation factor, the global augmentation search of the solution space for the second target individual is as follows:
[0075] in, Indicates the first t During the training process, the first j A second target individual, Indicates the first jThe second target individual after a global augmented search. Represents the optimal individual. This represents a random number between (-0.1, 0.1). This represents the mean value obtained from all individuals targeting the second objective; that is, each parameter dimension is the mean value of the parameter corresponding to all individuals targeting the second objective in the same dimension. Represents a random number between (0, 1); Determine whether the fitness of the second target individual after the global augmentation search has increased. If so, use the second target individual after the global augmentation search as the third target individual; otherwise, directly use the original second target individual as the third target individual.
[0076] This technical solution achieves a highly efficient combination of local development and global exploration through the synergy of balanced spiral search and greedy quantum search, resulting in significant technical improvements. The balanced spiral search strategy simulates the spiral approach path of biological predation, performing refined and multi-level local mining of the neighborhood of the current optimal solution, effectively improving the convergence accuracy and solution quality of the algorithm. The greedy quantum search strategy, on the other hand, introduces probabilistic perturbations based on quantum mechanics. By incorporating population mean information and an adaptive decay factor, it guides individuals to escape local optima and make quantum leaps to a better global region. Its core greedy selection mechanism ensures the non-degradation of population quality in each iteration, guaranteeing the stability of the algorithm. The combination of these two approaches allows the algorithm to both deeply explore local potential and overcome global limitations, thus finding the optimal solution faster and more accurately, significantly improving the performance and training efficiency of the final model.
[0077] Traditional gradient descent methods converge slowly when dealing with high-dimensional, non-convex parameter spaces in deep learning models. The fuzzy quantum deep learning algorithm of this invention is a swarm intelligence optimization algorithm that, through a parallelized population search mechanism, can explore multiple regions simultaneously. The design of fuzzy search and spiral search strategies gives the algorithm strong global search capabilities in the early stages of training, quickly locating high-quality solution regions; in the later stages of training, it focuses on fine-grained local searches, accelerating convergence. This adaptive search balancing mechanism significantly shortens the time required for the model to reach stability and high accuracy, improving the deployment efficiency of the maintenance and inspection system.
[0078] like Figure 2 As shown, this application embodiment provides an intelligent equipment maintenance and inspection device, including: The data acquisition module 201 is used to construct a digital twin model corresponding to the target device and detect the target operation data corresponding to the target device through the digital twin model corresponding to the target device. Data integration module 202 is used to obtain the real fault labels corresponding to the target operation data input by the staff, and to jointly construct prior experience data with the target operation data and the corresponding real fault labels; Deep learning module 203 is used to learn the data relationships in the prior experience data using a fuzzy quantum deep learning algorithm to obtain an intelligent maintenance and inspection model for the equipment; The intelligent maintenance module 204 is used to detect the corresponding real-time operating data of the target device through the digital twin model, and call the intelligent maintenance model of the device to identify the real-time operating data to determine the intelligent maintenance identification result of the device. The report generation module 205 is used to generate an intelligent equipment maintenance report based on the intelligent equipment maintenance identification results, and push the intelligent equipment maintenance report to the equipment designated by the staff.
[0079] The intelligent equipment maintenance and inspection device provided in this application embodiment can execute the above-mentioned method and technical solution. Its principle and beneficial effects are similar, and will not be described again here.
[0080] like Figure 3 As shown in the figure, this application embodiment provides an intelligent equipment maintenance and inspection device, which may include a memory 301 and a processor 302. Exemplarily, the memory 301 and the processor 302 are interconnected via a bus 303.
[0081] The memory 301 stores computer-executed instructions; The processor 302 executes the computer execution instructions stored in the memory, causing the processor to perform any of the intelligent equipment maintenance and inspection methods described above.
[0082] Figure 3 The intelligent maintenance and inspection equipment shown in the embodiment can execute the technical solution shown in the above method embodiment. Its implementation principle and beneficial effects are similar, and will not be repeated here.
[0083] This application also provides a computer-readable storage medium storing a computer program / instructions thereon, which, when executed by a processor, implements the steps in the intelligent equipment maintenance and inspection method disclosed in this application.
[0084] This application also provides a computer program product that, when run on an electronic device, enables the processor to execute the steps in the intelligent equipment maintenance and inspection method disclosed in this application.
[0085] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.
[0086] This application describes embodiments with reference to flowchart illustrations and / or block diagrams of methods, apparatuses, electronic devices, and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing terminal device to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing terminal device, generate instructions for implementing the flowchart... Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0087] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing terminal device to operate in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0088] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal equipment, causing a series of operational steps to be performed on the computer or other programmable terminal equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable terminal equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0089] Although preferred embodiments of the present application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of the embodiments of the present application.
[0090] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or terminal device that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or terminal device. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or terminal device that includes said element.
[0091] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.
Claims
1. A method for intelligent maintenance and inspection of equipment, characterized in that, include: Construct a digital twin model corresponding to the target device, and detect its corresponding target operation data through the digital twin model of the target device; Obtain the actual fault labels corresponding to the target operating data input by the staff, and construct prior experience data by combining the target operating data and the corresponding actual fault labels; The fuzzy quantum deep learning algorithm is used to learn the data relationships in the prior experience data to obtain an intelligent maintenance and inspection model for the equipment; The real-time operating data of the target device is detected by the digital twin model corresponding to the target device, and the intelligent maintenance and inspection model of the device is called to identify the real-time operating data to determine the intelligent maintenance and inspection identification result of the device. Based on the intelligent maintenance and inspection results, an intelligent maintenance and inspection report is generated for the equipment, and the report is pushed to the equipment designated by the staff.
2. The intelligent equipment maintenance method according to claim 1, characterized in that, The target operating data includes at least the vibration signal, temperature signal, current signal, and voltage signal of the target device at multiple consecutive sampling time points.
3. The intelligent equipment maintenance method according to claim 1, characterized in that, The data relationships in the prior experience data are learned using a fuzzy quantum deep learning algorithm to obtain an intelligent maintenance and inspection model for the equipment, including: A target deep learning model is constructed using a CNN-LSTM model, and the model parameters of the target deep learning model are initialized to obtain the training population; For any individual in the training population, the fitness of the individual is obtained based on the prior experience data, and the optimal individual is determined based on the fitness. For any given individual, based on the optimal individual, and using a fuzzy search strategy, a fuzzy search of the solution space is performed to obtain the first target individual; For any first target individual, a balanced spiral search strategy is used to perform a local augmentation search in the solution space of the first target individual to obtain the second target individual; For any number of second target individuals, a greedy quantum search strategy is used to perform a global augmentation search of the solution space of the second target individuals to obtain the third target individuals; Obtain the current number of training iterations, and obtain the training status based on the current number of training iterations; wherein, the training status includes a training completed status or a training incomplete status; When the training state is the training completed state, the final parameters of the target deep learning model are determined according to the third target individual to obtain the intelligent maintenance and inspection model of the equipment. When the training state is incomplete, the process returns to the step of determining the optimal individual based on the third target individual, thus entering the next training process.
4. The intelligent equipment maintenance method according to claim 3, characterized in that, Initializing the model parameters of the target deep learning model to obtain a training population includes: initializing the model parameters of the target deep learning model using a random initialization strategy or a chaotic mapping initialization strategy to obtain a training population containing multiple individuals; wherein, the individuals include some or all of the model parameters to be trained of the target deep learning model.
5. The intelligent equipment maintenance method according to claim 3, characterized in that, The process of obtaining the fitness of an individual based on the prior experience data and determining the optimal individual based on the fitness includes: For any individual, the model parameters contained in the individual are applied to the target deep learning model to obtain the target deep learning model after applying the parameters; The target running data in the prior experience data is used as the actual input of the target deep learning model after applying the parameters, and the real fault labels in the prior experience data are used as the expected output of the target deep learning model after applying the parameters to obtain the cross-entropy loss function value. The cross-entropy loss function value is added to a preset constant term to obtain a non-zero data term, and the reciprocal of the non-zero data term is taken to obtain the fitness of the individual. Obtain the fitness of all individuals and determine the individual with the highest fitness as the optimal individual.
6. The intelligent equipment maintenance method according to claim 3, characterized in that, Based on the optimal individual, and using a fuzzy search strategy to perform a fuzzy search of the solution space, the first target individual is obtained, including: Based on the fitness of the individual, the first fuzzification factor and the second fuzzification factor are obtained as follows: in, Indicates the first i The first fuzziness factor for each individual Indicates the first i The second fuzzification factor for each individual, Indicates the first i The fitness of an individual i =1,2,…,MP, where MP represents the total number of individuals. This represents the fitness of the optimal individual. This represents the fitness of the worst individual, which is the individual with the lowest fitness. a This represents the constant term in [0.8, 1). Based on the first fuzzification factor and the second fuzzification factor, the fuzzification control coefficients corresponding to the training population are obtained as follows: in, Represents the fuzzy control coefficient. express and The maximum value in; Based on the fuzzification control coefficients, the first search factor and the second search factor are obtained as follows: in, Indicates the first search factor. Indicates the second search factor. This represents the first random number, which is either 0 or 1. This represents a second random number that is either 0 or 1. Indicates the maximum number of training iterations; Based on the optimal individual, the first search factor, and the second search factor, a fuzzy search of the solution space is performed on the individual to obtain the first target individual: in, Indicates the first t During the training process, the first i Individual, Indicates the first i The first target individual, Represents the optimal individual. Represents the natural constant. This means that among all individuals, except for The first random individual outside of [the previous one].
7. The intelligent equipment maintenance method according to claim 6, characterized in that, A balanced spiral search strategy is used to perform a local augmentation search of the solution space on the first target individual to obtain the second target individual, including: The spiral search factor is obtained as follows: in, 'b' represents the spiral search factor, and 'b' represents the spiral path parameter. l The helix step size parameter is represented by cos, and cos represents the cosine function. This represents a third random number between (0,1). Represents the inverse hyperbolic tangent function; Based on the spiral search factor, a local enhancement search in the solution space is performed on the first target individual to obtain the second target individual: in, Indicates the first t During the training process, the first j The first target individual, Indicates the first j A second target individual, Indicates the second search factor. This means that among all individuals in the first target category, excluding A second random individual other than; This represents the fourth random number, which is either 0 or 1.
8. The intelligent equipment maintenance method according to claim 7, characterized in that, A greedy quantum search strategy is used to perform a global augmentation search of the solution space on the second target individual to obtain the third target individual, including: The attenuation factor is obtained as follows: in, The attenuation factor is represented by sine, and sin represents the sine function. Let represent the fitness of the best individual during the (t-1)th training iteration. Represents pi; Based on the attenuation factor, the global augmentation search of the solution space for the second target individual is as follows: in, Indicates the first t During the training process, the first j A second target individual, Indicates the first j The second target individual after a global augmented search. Represents the optimal individual. This represents a random number between (-0.1, 0.1). This represents the mean individual obtained from all individuals with the second objective. Represents a random number between (0, 1); Determine whether the fitness of the second target individual after the global augmentation search has increased. If so, use the second target individual after the global augmentation search as the third target individual; otherwise, directly use the original second target individual as the third target individual.
9. An intelligent equipment maintenance and inspection device, characterized in that, include: The data acquisition module is used to construct a digital twin model corresponding to the target device and to detect the target operating data corresponding to the target device through the digital twin model corresponding to the target device. The data integration module is used to obtain the real fault labels corresponding to the target operation data input by the staff, and to jointly construct prior experience data by the target operation data and its corresponding real fault labels. The deep learning module is used to learn the data relationships in the prior experience data using the fuzzy quantum deep learning algorithm to obtain an intelligent maintenance and inspection model for the equipment. The intelligent maintenance module is used to detect the corresponding real-time operating data of the target device through the digital twin model, and call the intelligent maintenance model of the device to identify the real-time operating data to determine the intelligent maintenance identification result of the device. The report generation module is used to generate an intelligent equipment maintenance report based on the intelligent equipment maintenance identification results, and push the intelligent equipment maintenance report to the equipment designated by the staff.
10. An intelligent equipment maintenance and inspection device, characterized in that, Including processor and memory; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the intelligent equipment maintenance and inspection method as described in any one of claims 1 to 8.