Intelligent Equipment Operation and Maintenance Methods and Systems Based on Industrial Big Data
By collecting multidimensional feature data and physical models of equipment degradation, and combining residual calculation and constraint logic programming, the problems of false alarms and false alarms in equipment health assessment were solved. This enabled automatic assessment of equipment health status and analysis of fault propagation chains, improving the automation of maintenance decisions and resource optimization, and increasing maintenance efficiency.
Patent Information
- Application Number
- CN202611239684.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-17
- Publication Date
- 2026-11-10
AI Technical Summary
Existing technologies suffer from frequent false alarms and missed alarms in equipment health assessments. Furthermore, fault location and maintenance decisions rely on human experience and lack systematic propagation reasoning and optimization under resource constraints, resulting in low maintenance efficiency.
By collecting multidimensional feature data, combining it with equipment degradation physical models and relationship graphs, and using residual calculation and constraint logic programming to generate optimal maintenance solutions, the system can automatically assess the health status of equipment and analyze fault propagation chains, thereby optimizing maintenance decisions.
It improves the accuracy of equipment health assessment, reduces false alarm rates, and automates and optimizes maintenance decisions, thereby reducing unplanned equipment downtime and maintenance costs.
Smart Images

Figure CN123070922A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of intelligent operation and maintenance technology for industrial equipment, specifically to intelligent operation and maintenance methods and systems for equipment based on industrial big data. Background Technology
[0002] Industrial equipment maintenance has long relied on single-signal threshold alarms and periodic inspection strategies. On-site, fixed upper limits are typically set for vibration amplitude or temperature values; once the monitored signal exceeds the threshold, an alarm is triggered, and maintenance personnel troubleshoot and schedule repairs based on experience. This approach can be effective in scenarios with simple equipment degradation mechanisms and stable operating conditions, but its limitations become increasingly apparent when facing modern complex electromechanical systems. The shortcomings of existing technical solutions are mainly reflected in two aspects. First, health assessment methods based on single signal thresholds struggle to capture the multidimensional characteristics of equipment degradation, leading to frequent false alarms and missed alarms. Anomalies in different physical quantities such as vibration, temperature, and current are often coupled; relying solely on independent thresholds ignores the correlation between signals and fails to reflect the true state of the internal physical degradation process of the equipment. Some methods attempt to introduce data-driven models, but purely data-driven approaches lack an expression of the physical failure mechanisms of equipment, exhibit poor generalization ability when samples are insufficient or operating conditions change, and show significant deviations between health assessment results and actual degradation trends.
[0003] On the other hand, fault location and maintenance decisions heavily rely on human experience, lacking systematic propagation reasoning and optimization mechanisms under resource constraints. When equipment health declines, maintenance personnel must manually consult drawings and historical records to deduce the root cause of the fault. For complex equipment groups with interconnected relationships, manual analysis struggles to fully trace the fault propagation path, easily overlooking key nodes or mislocating the fault. Maintenance plans typically only consider a single fault point, ignoring various practical constraints such as total maintenance cost, spare parts inventory, personnel qualifications, and maintenance time windows. This leads to frequent resource conflicts, project delays, or repeated downtime during maintenance execution, making it difficult for overall operation and maintenance efficiency to meet the needs of refined management.
[0004] To address these issues, a health assessment method is needed that can integrate multi-source data with the physical degradation mechanism of equipment, accurately depicting the trend of equipment status changes while reducing false alarms. Furthermore, a method is needed that can automatically analyze fault propagation chains and generate optimal maintenance plans under complex resource constraints, replacing the current reliance on manual fault location and planning, thereby reducing unplanned equipment downtime and controlling maintenance costs. Summary of the Invention
[0005] This paper presents an intelligent equipment operation and maintenance method and system based on industrial big data. It utilizes multi-dimensional features and residual calculation of equipment degradation physical models to achieve reliable assessment of equipment health status. It also combines equipment relationship graph analysis to resolve fault propagation chains and solves the optimal maintenance plan under resource constraints through constraint logic programming. This improves the accuracy of health assessment, reduces the false alarm rate, and automates and optimizes maintenance decisions.
[0006] To achieve the above objectives, the present invention provides the following technical solution: The present invention provides an intelligent equipment operation and maintenance method based on industrial big data, comprising: collecting vibration time-series signals, temperature change sequences, and current waveform data during equipment operation; performing Fourier transform on the vibration time-series signals to extract spectral features, performing differential calculation on the temperature change sequences to extract the temperature change rate, and extracting harmonic distortion from the current waveform data; combining the spectral features, temperature change rate, and harmonic distortion into a multi-dimensional feature vector; calculating the residual between the multi-dimensional feature vector and the equipment degradation physical model to obtain an equipment health index; triggering a maintenance request when the health index is lower than a preset threshold; based on the maintenance request, parsing the fault propagation chain from the equipment relationship graph to generate a list of suspected fault nodes; solving for the optimal maintenance solution using constraint logic programming based on the list of suspected fault nodes and maintenance history records; and outputting the optimal maintenance solution to the maintenance management system.
[0007] By simultaneously acquiring multi-source data such as vibration, temperature, and current, and extracting spectral features, temperature change rate, and harmonic distortion, the system can comprehensively reflect the equipment status from the perspectives of frequency domain, time domain, and waveform distortion, overcoming the limitation of incomplete fault characterization by a single signal. The residual calculation between the multi-dimensional feature vector and the equipment degradation physical model quantifies the deviation between actual characteristics and the ideal degradation state using Euclidean distance, and smooths it using an exponentially weighted moving average. This ensures that the obtained health index retains sensitivity to early degradation while effectively suppressing transient noise interference, improving the accuracy and robustness of alarms. When the health index falls below a threshold, a reverse breadth-first search is performed using an equipment relationship graph with equipment components as nodes and physical connections or process dependencies as directed edges. This analyzes the fault propagation chain and filters nodes with high influence in the fault propagation path by calculating betweenness centrality. Simultaneously, fault nodes from maintenance history are integrated to generate a list of suspected fault nodes, thereby narrowing the investigation scope and improving the targeting and coverage of fault location. Building upon this foundation, constraint logic programming (CLP) technology is introduced to abstract maintenance plans into binary decision variables. This comprehensively considers multi-dimensional constraints such as the upper limit of total maintenance cost, the number of available maintenance personnel, mandatory maintenance at critical nodes, spare parts inventory, scheduling conflicts for certified maintenance personnel, and maintenance windows determined by production scheduling. The optimization objective is to minimize the weighted sum of equipment downtime and maintenance costs. A solver searches for the globally optimal maintenance plan in the feasible solution space, achieving automatic optimization of maintenance decisions under complex resource constraints. Finally, the optimal maintenance plan is sequentially broken down into a series of work orders. Each work order includes the maintenance object, operation code, spare parts list, and estimated working hours. These work orders are pushed to the computerized maintenance management system via a standard interface, triggering work order creation and resource reservation. Confirmation receipts and timeout resending mechanisms are also implemented to ensure reliable issuance of maintenance instructions and form a closed-loop management system, thereby significantly improving the intelligence level and execution efficiency of industrial equipment operation and maintenance.
[0008] Preferably, when collecting equipment operation data, accelerometers, thermocouples, and current transformers are installed at key measuring points of the equipment. Data is collected synchronously at a fixed sampling frequency. The raw data is segmented using a sliding window, and outliers exceeding three standard deviations within the window are removed. The data after removing outliers are then aligned by timestamps to generate continuous vibration time-series signals, temperature change sequences, and current waveform data. This eliminates the interference of occasional sensor anomalies and data asynchrony on subsequent analysis, ensuring the accuracy and consistency of feature extraction input. Preferably, when extracting spectral features, a Hanning window function is applied to the vibration time-series signal, followed by a Fast Fourier Transform to obtain the amplitude spectrum. The frequencies and amplitudes corresponding to the first five peak values are extracted from the amplitude spectrum to form a vibration spectrum feature vector, which is then normalized to a unit length. This method can suppress spectral leakage while highlighting the main vibration modes, enhancing the feature's ability to distinguish different fault modes. In calculating health indicators, a physical model of equipment degradation is pre-established, using multi-dimensional feature vectors as input and health indicators as output. The model parameters are calibrated by finite element simulation, giving the residual calculations a clear physical meaning and thus more accurately reflecting the actual degradation of the equipment. When analyzing fault propagation chains, all reachable nodes obtained through a reverse breadth-first search starting from the equipment component triggering the maintenance request are considered candidate fault nodes. The betweenness centrality of each candidate fault node is calculated, and high-impact nodes with a pre-set proportion are selected in descending order. This is then combined with the nodes that have experienced faults in the maintenance history to generate the final list of suspected fault nodes. This approach considers both the propagation criticality of the topology and historical fault experience, reducing the risk of missed detections. When solving for the optimal maintenance plan using constraint logic programming, constraints on the quantity of repair parts in the enterprise resource planning system and the scheduling constraints for certified personnel are obtained from the human resources system. Maintenance window time constraints are set according to the production scheduling plan to ensure that the generated optimal maintenance plan is actually executable under realistic resource conditions, avoiding the generation of infeasible maintenance plans.
[0009] This invention also provides an intelligent equipment operation and maintenance system based on industrial big data, including a processor and a memory. The memory stores a program, and when the processor executes the program, it implements the aforementioned intelligent equipment operation and maintenance method based on industrial big data. This system possesses the same technical effects as the method described above, capable of online acquisition of multi-source signals, automatic assessment of equipment health status, and generation of optimal maintenance solutions under resource constraints in the event of anomalies. It also transmits maintenance instructions in a closed loop to the maintenance management system, making it suitable for predictive maintenance and intelligent decision-making in complex industrial scenarios.
[0010] The technical effects and advantages provided by the present invention in the above technical solution are as follows:
[0011] Spectral features, temperature change rates, and harmonic distortion are extracted from vibration time-series signals, temperature change sequences, and current waveform data, and combined to form a multi-dimensional feature vector. Based on this, the Euclidean distance between this multi-dimensional feature vector and the predicted vector output by the equipment degradation physical model is calculated, and an exponentially weighted moving average is applied to this distance to generate an equipment health index. The equipment degradation physical model is calibrated through finite element simulation and incorporates the mechanical and thermal degradation laws between equipment materials, structure, and operating conditions. The residual between the multi-dimensional feature vector and the physical model directly reflects the degree of deviation between the current observed data and the expected degradation trajectory. While retaining data-driven sensitivity, the physical model constraint suppresses spurious fluctuations introduced by noise or transient operating condition disturbances. The health index can reflect early and more smoothly the early weak degradation trend of the equipment, and maintains a low false alarm rate even under low load or variable operating conditions. When a health index falls below a preset threshold, triggering a maintenance request, the fault propagation chain is analyzed from the equipment relationship graph. Starting from the equipment component that triggered the request, a width-first search is performed along the directed edges to obtain candidate fault nodes. High-impact nodes are then selected using betweenness centrality calculation, and their union with nodes that have previously experienced faults in the maintenance history is taken to generate a list of suspected fault nodes. The maintenance plan is defined as a binary decision variable for each suspected node. The constraint set includes a maximum total maintenance cost, a limit on the number of available maintenance personnel, a hard constraint requiring maintenance of critical nodes, maintenance parts inventory constraints obtained from the enterprise resource planning system, certified personnel scheduling constraints obtained from the human resources system, and maintenance window constraints defined by the production scheduling plan. The optimization objective is to minimize the weighted sum of equipment downtime and maintenance costs. A constraint logic solver searches the feasible solution space for the maintenance plan with the minimum objective function value. The fault propagation chain analysis utilizes the physical connections and process dependencies in the equipment relationship graph, making the generation of suspected fault nodes more physically interpretable. The union formed by combining maintenance history can take into account both high-impact nodes and historically weak nodes, avoiding missed detections. When faced with multiple resource coupling constraints, constraint logic programming can simultaneously handle hard constraints and linear weighted optimization objectives. The resulting optimal maintenance plan is directly executable in terms of spare parts availability, personnel qualifications, and time windows. It eliminates resource conflicts that are common when manually formulating maintenance plans, reduces overall equipment downtime, and controls total maintenance costs. Attached Figure Description
[0012] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.
[0013] Figure 1This is a flowchart of an intelligent equipment operation and maintenance method based on industrial big data.
[0014] Figure 2 This is a flowchart of the process for generating suspected fault nodes based on reverse breadth-first search and betweenness centrality analysis;
[0015] Figure 3 This is a schematic diagram of the spectrum of the vibration time-series signal and the peak frequencies of the first five orders;
[0016] Figure 4 This is a scatter plot showing the trade-off between downtime and cost for maintenance solutions.
[0017] Figure 5 This is a time sequence diagram of the optimal maintenance plan work order execution. Detailed Implementation
[0018] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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 embodiments of the present invention, 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.
[0019] See Figure 1 This invention provides an intelligent equipment operation and maintenance method based on industrial big data, comprising the following steps: collecting vibration time-series signals, temperature change sequences, and current waveform data during equipment operation; performing Fourier transform on the vibration time-series signals to extract spectral features, performing differential calculation on the temperature change sequences to extract the temperature change rate, and extracting harmonic distortion from the current waveform data; combining the spectral features, temperature change rate, and harmonic distortion into a multi-dimensional feature vector; calculating the residual between the multi-dimensional feature vector and the equipment degradation physical model to obtain an equipment health index; triggering a maintenance request when the health index is lower than a preset threshold; based on the maintenance request, parsing the fault propagation chain from the equipment relationship graph to generate a list of suspected fault nodes; solving for the optimal maintenance solution using constraint logic programming based on the list of suspected fault nodes and maintenance history records; and outputting the optimal maintenance solution to the maintenance management system.
[0020] Example 1:
[0021] In practice, the method for collecting vibration timing signals, temperature change sequences, and current waveform data during equipment operation is as follows: Accelerometers, thermocouples, and current transformers are installed at key measuring points on the equipment, and data is collected synchronously at a fixed sampling frequency. The key measuring points are determined based on the equipment's structural mechanics analysis and historical high-fault locations. The fixed sampling frequency is set according to the equipment's operating speed and signal analysis requirements; for example, it may be set to collect frequency values corresponding to 128 points per revolution. Synchronous acquisition is achieved through a unified time-synchronized acquisition card, ensuring that the data output from the accelerometers, thermocouples, and current transformers have the same timestamp.
[0022] The collected raw data is processed in segments using a sliding window. The window length is set to the number of data points in one complete operating cycle of the equipment, and the sliding step size is set to half the window length. Within each sliding window, the mean and standard deviation of the data within the window are calculated, and outliers exceeding three times the standard deviation are removed. After removing outliers, missing data positions are filled using linear interpolation. The acceleration, temperature, and current data, after outlier removal, are aligned by timestamps to generate continuous vibration time-series signals, temperature change sequences, and current waveform data. The vibration time-series signal represents the vibration acceleration amplitude varying over time, the temperature change sequence represents the temperature values varying over time, and the current waveform data represents the instantaneous current values varying over time.
[0023] The specific implementation of Fourier transform to extract spectral features from vibration time-series signals is as follows: After applying a Hanning window function to the vibration time-series signal, a Fast Fourier Transform is performed to obtain the amplitude spectrum. The expression for the Hanning window function is: ,in The sampling point number, The length of the Hanning window represents the number of sample points participating in the Fast Fourier Transform. Applying the Hanning window function reduces spectral leakage. The frequencies and amplitudes corresponding to the first five peaks are extracted from the amplitude spectrum to form a vibrational spectrum feature vector. The first five peaks refer to the five local maxima with the largest amplitudes in the amplitude spectrum. Local maxima are determined as follows: if the amplitude of a frequency point is greater than the amplitudes of its two immediately preceding and following frequencies, then that frequency point is a local maximum. The five extracted frequency values are denoted as follows: , , , , The corresponding amplitudes are denoted as follows: , , , , The resulting vibration spectrum feature vector is The method to normalize the vibration spectrum eigenvector to a unit length is as follows: calculate the Euclidean norm of the vibration spectrum eigenvector. The normalized spectral characteristics are ,in This represents the normalized spectral characteristics. The Euclidean norm represents the eigenvector of the vibration spectrum. The normalized eigenvector has a length of 1, eliminating the influence of the absolute magnitude of the signal amplitude on subsequent analysis. The rate of temperature change is obtained by performing a first-order backward difference on the temperature change sequence. Harmonic distortion is obtained by calculating the total harmonic distortion rate from the current waveform data; the total harmonic distortion rate is the ratio of the square root of the sum of the squares of the effective values of each harmonic current to the effective value of the fundamental current.
[0024] See Figure 3 In the graph, the horizontal axis represents frequency in Hertz (Hz), ranging from 0 to 5000 Hz, and the vertical axis represents vibration amplitude in meters per second (m / s), ranging from 0 to approximately 2.5 m / s. The blue curve represents the amplitude spectrum of the vibration signal, reflecting the amplitude distribution of vibration acceleration across various frequency components during equipment operation. The red dots mark the first five peak frequencies and their corresponding amplitudes in the amplitude spectrum. Specifically, the peaks are, in order, the first peak with the highest amplitude of approximately 2.36 m / s, and the second to fifth peaks, decreasing sequentially. All amplitudes are greater than those of nearby frequencies, satisfying the criteria for local maxima.
[0025] As described in Example 1, the vibration timing signal is processed by the Hanning window function and then subjected to a Fast Fourier Transform to obtain the amplitude spectrum. The five peak frequencies are indicated by red dots. to and amplitude to This forms the vibration spectrum feature vector. The peak values in the figure are mainly concentrated in the low-frequency range of 0 to 500 Hz, and the amplitude decreases as the frequency increases, indicating that the main energy of the equipment vibration signal is concentrated in the lower frequency range, consistent with the characteristics of normal equipment operation. The first five peak frequencies with higher amplitudes provide key inputs for subsequent feature normalization and equipment health assessment.
[0026] Example 2:
[0027] In practical implementation, the method for obtaining the equipment health index based on the residual calculation between the multidimensional feature vector and the equipment degradation physical model is as follows: A physical model of equipment degradation is pre-established, taking the multidimensional feature vector as input and the health index as output. The physical model of equipment degradation takes the following form:
[0028]
[0029] in, Indicates time The predicted health index output by the physical model of equipment degradation. This represents the initial health baseline value. The results were obtained through finite element simulation under the condition that the equipment was in brand new factory condition and operating under standard working conditions. The value is set as the maximum health score in the finite element simulation model. Set as . It represents the natural constant, that is, the base of the natural logarithm. Represents the degradation rate coefficient. The range of values is greater than real numbers, The degradation process is obtained by finite element simulation, which simulates the degradation process of a device from a brand-new state to a given degradation state, and by fitting parameters to the simulated degradation data. Indicates time The multidimensional feature vectors obtained by collecting and combining them This represents the standard multidimensional feature vector reference value when the equipment is operating under standard conditions in a brand-new factory state. The values of each dimension are determined by the simulation output of finite element simulation under the condition of no degradation of the equipment. Representing multidimensional feature vectors Compared with standard multidimensional feature vector reference value The Euclidean norm between two vectors is the square root of the sum of the squares of the differences between the two vectors along their corresponding dimensions.
[0030] The model parameters of the physical model of equipment degradation are calibrated by finite element simulation. The calibration process is as follows: a finite element model is established using the three-dimensional geometric model of the equipment and material physical parameters. Different degradation conditions are set in the finite element model. Vibration response, temperature field distribution, and current signal waveforms at measuring points on the equipment surface under different degradation conditions are obtained through transient simulation. The simulation results are used to extract spectral features, temperature change rate, and harmonic distortion using the same feature extraction method as actual acquisition, and combined into a simulation multidimensional feature vector. The simulation multidimensional feature vector is then fitted with the set degradation degree label through regression to obtain the degradation rate coefficient. The specific value.
[0031] Calculate the Euclidean distance between the current multidimensional feature vector and the predicted vector output by the equipment degradation physical model under the same conditions. The same conditions refer to the operating parameters at the time corresponding to the current multidimensional feature vector, including equipment load rate, ambient temperature, and operating speed. The equipment degradation physical model uses the current multidimensional feature vector and operating parameters as input to generate a predicted health index. The Euclidean distance is calculated as follows: ,in: Represents Euclidean distance. This represents the actual health projection value directly calculated from the current multidimensional feature vector. This is obtained by substituting the current multidimensional feature vector into the inverse mapping function of the device degradation physical model. This represents the predicted health index output by the physical model of equipment degradation under the same operating parameters.
[0032] An exponentially weighted moving average is applied to the Euclidean distance to generate a smoothed equipment health index. The smoothing coefficient for the exponentially weighted moving average is [not specified]. Set as , The range of values is to between. Set as The basis for this is that when the equipment is operating under steady-state conditions, the instantaneous fluctuations in the health index are mainly caused by measurement noise, and the smoothing coefficient... Taking a smaller value allows historical smoothed values to have a larger weight, thus suppressing the interference of instantaneous fluctuations on the judgment of health trend. The recursive calculation method of the exponentially weighted moving average is: taking the current time... The Euclidean distance is denoted as The previous moment The smoothness health index is marked as Then at the current moment Smoothed device health index Satisfying the recurrence relation ,in Indicates the current time Smoothed equipment health indicators Represents the smoothing coefficient. Indicates the current time The calculated Euclidean distance, Indicates the previous moment Smoothed equipment health index. Initial value processed by exponential weighted moving average. It is set to the first Euclidean distance value collected since the device's most recent maintenance.
[0033] Example 3:
[0034] In specific implementation, please refer to Figure 2The method for parsing fault propagation chains from equipment relationship graphs is as follows: The equipment relationship graph is read, with equipment components as nodes and physical connections or process dependencies as directed edges. The equipment relationship graph is stored in a graph database. Each equipment component node in the graph has a unique identifier and a component type attribute. Each directed edge has a connection type attribute, which can be either a physical connection or a process dependency. A physical connection indicates a mechanical fastening, pipeline connection, or electrical cable connection between two equipment components. A process dependency indicates that the output materials or energy of the upstream equipment component are directly supplied to the downstream equipment component.
[0035] Starting with the device component that triggered the maintenance request, a reverse breadth-first search is performed along the directed edges to obtain all reachable nodes as candidate fault nodes. The reverse breadth-first search refers to traversing in the opposite direction of the directed edges; that is, starting from the starting point, visiting all neighboring nodes connected to the starting point via incoming edges, and then expanding outwards layer by layer. During the search, a set of visited nodes is maintained to avoid repeated traversals, and the search terminates when no new reachable nodes are found. All reachable nodes obtained from the search constitute the candidate fault node set.
[0036] Calculate the betweenness centrality of each candidate faulty node, and select nodes with betweenness centrality exceeding a preset threshold to form a list of suspected faulty nodes. The formula for calculating betweenness centrality is:
[0037]
[0038] in, Indicates candidate fault nodes betweenness centrality and This represents any two distinct equipment component nodes in the equipment relationship diagram, and and None of them are equal to , Represents a node With nodes The number of all shortest paths between them. Indicates at node With nodes The nodes that are passed through in all shortest paths between them The number of shortest paths. Summation symbol. This represents all devices in the device relationship graph that satisfy... Not equal to , Not equal to and Not equal to node pairs Accumulate.
[0039] The ratio of the number of shortest paths passing through candidate faulty nodes to the total number of shortest paths is used as the betweenness centrality. Specifically, for each pair of nodes in the device relationship graph... Calculate all shortest paths using either the Floyd-Warshall algorithm or the Dijkstra algorithm, and then count the total number of shortest paths. and among them, passing through candidate fault nodes Number of shortest paths , the ratio Accumulate to obtain candidate faulty nodes. betweenness centrality .
[0040] After obtaining the betweenness centrality of each candidate faulty node, the nodes are sorted from high to low, and the top 30% are selected as high-influence nodes. In specific implementation, the total number of candidate faulty nodes is obtained. Calculate the number of high-impact nodes ,in This indicates rounding up, sorting by betweenness centrality. Each node is marked as a high-impact node.
[0041] A list of potential fault nodes is generated by taking the union of high-impact nodes and nodes that have previously failed in the maintenance history. The maintenance history is stored in the maintenance management database and includes the fault occurrence time, faulty equipment / component identifier, and fault description. All equipment / component identifiers that have failed in the maintenance history are extracted to form a set of historical fault nodes. The set of high-impact nodes and the set of historical fault nodes are then merged and deduplicated to obtain the list of potential fault nodes. Each entry in the list of potential fault nodes includes an equipment / component identifier and component type, which serve as input for subsequent solutions to the optimal maintenance plan.
[0042] Example 4:
[0043] In practice, the method of solving the optimal maintenance plan using constraint logic programming is as follows: the maintenance plan variable is defined as a binary decision variable, representing whether to perform maintenance on each node in the suspected fault node list. The suspected fault node list contains... The node, the node in the list of suspected faulty nodes Each node defines a binary decision variable. ,in The value is from arrive Integers. Indicates the first Each node performs maintenance operations. Indicates no Each node performs maintenance operations.
[0044] Define a set of constraints, including a maximum total maintenance cost, a limit on the number of available maintenance personnel, and hard constraints requiring critical nodes to be repaired. The maximum total maintenance cost constraint is set as follows: for the node in the list of suspected faulty nodes... For each node, retrieve the standard maintenance cost corresponding to that node from the maintenance cost database. Standard maintenance cost Including the cost of replacement parts and standard labor costs, the maximum total repair cost is [amount missing]. Based on the current budget balance obtained from the Enterprise Resource Planning (ERP) system, the total maintenance cost constraint is: ,in This represents the total number of nodes in the list of suspected faulty nodes. Indicates the first Standard maintenance cost per node Indicates the first The binary decision variables corresponding to each node This indicates the upper limit of the total maintenance cost. The method for setting the limit on the number of available maintenance personnel is: obtain the total number of certified maintenance personnel available for maintenance work within the current shift period from the human resources system. Query for the first Number of maintenance personnel required to perform maintenance operations at each node The number of available maintenance personnel is constrained to: ,in This indicates the total number of certified maintenance personnel available for maintenance work. Indicates the first The number of maintenance personnel required to perform maintenance operations on each node. The hard constraint that critical nodes must be maintained refers to the requirement that equipment components marked as critical nodes in the equipment relationship graph must have maintenance operations included in the maintenance plan. This is represented by a set of indices corresponding to all critical nodes. The hard constraint of the critical node is for any , ,in Represents the set of key node indexes. Indicates belonging to The node index, Indicates key nodes The corresponding binary decision variables.
[0045] The constraint set also includes obtaining the current inventory quantity of repair parts from the enterprise resource planning system and setting a constraint that the quantity of consumable repair parts does not exceed the inventory quantity. For the first... For each node, retrieve the required materials (BOM) for repairing that node. Quantity of spare parts Obtain the first from the Enterprise Resource Planning system Current inventory quantity of spare parts The inventory constraint for maintenance consumables is for the first... Spare parts, ,in Indicates the spare parts type index. Indicates the repair of the first The first node requires the consumption of the first... The quantity of various spare parts Indicates the first The current inventory quantity of each spare part. Obtain the shift schedule of certified personnel for each maintenance position from the human resources system, and set a constraint that no more than one maintenance task from the same certified personnel can be performed within the same time period. Divide the maintenance time into... The time period, for the first The time period and the first The number of certified repair personnel, inquired about Has the certified maintenance worker been assigned a maintenance task during this time period? If so, is the maintenance task assigned to the first maintenance worker? The task of the node needs to use the first... A certified maintenance worker and the task was performed on If executed within a certain time period, it will occupy the identifier. Set as Otherwise set to Scheduling conflict constraints are for the first The time period and the first A certified repairman ,in Indicates a time period index. Indicates the total number of time periods. Index of certified maintenance personnel Indicates the repair of the first Is the node in the _th ... The time period occupies the first The system identifies the occupancy status of each certified maintenance worker. Maintenance windows are set according to the production scheduling plan, limiting the execution time of maintenance plans to within the designated windows. The start time of the permitted maintenance operation window is obtained from the production scheduling plan. and end time Repair No. The estimated time for each node is [time]. Maintenance task start time Must meet and .
[0046] The objective function, which aims to minimize the weighted sum of equipment downtime and maintenance costs, takes the following form:
[0047]
[0048] in, Indicates the weighting coefficient. The range of values is to Real numbers between. The settings are based on the criticality of the equipment in the production line; when the equipment is a bottleneck device in production... Values When the equipment is a non-bottleneck device Values Weighting coefficient The specific value is determined by the equipment's position in the production scheduling plan and the upstream and downstream buffer capacity. Indicates the repair of the first The estimated time for each node is... This represents the normalized reference value for downtime. The value represents the total estimated time required to repair all suspected faulty nodes. Indicates the first Standard maintenance cost per node This represents a normalized cost reference value. The value is the total standard maintenance cost corresponding to repairing all suspected faulty nodes. This represents the total number of nodes in the list of suspected faulty nodes. Indicates the first Each node corresponds to a binary decision variable.
[0049] The constraint logic solver is invoked to search the feasible solution space. The constraint logic solver uses a branch and bound algorithm to traverse the binary decision variables. The search process involves pruning branches that do not meet the constraint set and updating the bounds of feasible solutions that satisfy the constraint set based on the objective function value, until the search is completed or the preset search time limit is reached. The optimal maintenance solution is selected from the feasible solutions by choosing the maintenance plan with the smallest objective function value. The optimal maintenance plan includes a list of node indices for which maintenance operations should be performed and a corresponding time sequence for the maintenance tasks.
[0050] See Figure 4 In the figure, the horizontal axis represents the ratio of the downtime of the maintenance plan to the reference value, ranging from 0 to 1, and the vertical axis represents the ratio of the maintenance cost of the maintenance plan to the reference value, also ranging from 0 to 1. Blue dots represent the distribution of multiple feasible maintenance plans, and red stars indicate the optimal maintenance plan obtained through constraint logic programming.
[0051] The scatter plot distribution shows that feasible maintenance solutions are relatively dispersed within the two-dimensional space of downtime and maintenance cost, indicating a certain trade-off: solutions with shorter downtime may have higher maintenance costs, and vice versa. The overall trend suggests that as the downtime ratio increases, the distribution of the maintenance cost ratio becomes more widespread, and multiple feasible solutions with different combinations exist.
[0052] The optimal maintenance plan, marked with a red star, is located in the lower left corner of the figure. The ratio of downtime to the reference value is approximately 0.1, and the ratio of maintenance cost to the reference value is approximately 0.05, which is significantly lower than other feasible plans. This indicates that the plan achieves a weighted minimization of downtime and maintenance cost while meeting the constraints set in Example 4 (including the upper limit of maintenance cost, the number of available maintenance personnel, hard constraints on critical node maintenance, spare parts inventory limits, scheduling conflict limits, and maintenance window time, etc.).
[0053] The selection of the optimal solution conforms to the objective function through weight coefficients. The optimization objective is a weighted sum of downtime and maintenance costs. It takes into account the criticality of equipment in the production line, achieves rational resource allocation and maximizes maintenance efficiency, ensures that maintenance work is completed within a specified time window and minimizes production losses and maintenance costs.
[0054] Example 5:
[0055] In implementation, the optimal maintenance plan is broken down into a time-ordered sequence of work orders. Each work order includes the maintenance object, maintenance operation code, required spare parts list, and estimated working hours. The optimal maintenance plan contains a list of node indices for which maintenance operations should be performed and a corresponding maintenance task sequence, including the earliest start time and latest end time for each maintenance task. The breakdown process is as follows: traverse each maintenance node in the optimal maintenance plan, obtain the corresponding maintenance object name based on the equipment component identifier of the maintenance node, query the corresponding maintenance operation code from the standard maintenance operation code library based on the maintenance type of the maintenance node, generate a required spare parts list based on the bill of materials corresponding to the maintenance node, with each item in the required spare parts list including the spare parts material code, spare parts name, and required quantity, and calculate the estimated working hours based on the historical working hour statistics of the maintenance node. Combine the maintenance object, maintenance operation code, required spare parts list, and estimated working hours of each maintenance node into a work order, and sort all work orders in ascending order based on the earliest start time in the maintenance task sequence, forming a time-ordered sequence of work orders.
[0056] The work order sequence is pushed to the computerized maintenance management system via a standard interface, triggering the work order creation and resource reservation process. The standard interface uses an application programming interface based on a representational state transition architecture, and the interface data format uses JavaScript object representation. The push request contains an array of work order sequences, where each work order object includes a maintenance object field, a maintenance operation code field, a required spare parts list field, and an estimated working time field. Upon receiving the push request, the computerized maintenance management system parses each work order object in the work order sequence array, creates a work order record in its work order management module, and initializes the work order record's status to "pending dispatch." Simultaneously, the computerized maintenance management system's resource management module generates a spare parts reservation request based on the required spare parts list for each work order, locks the inventory quantity of the corresponding spare parts material code in the enterprise resource planning system, matches the skill tags of certified maintenance personnel based on the estimated working time and maintenance operation code of each work order, generates a personnel appointment request in the human resources management module, and marks the target time period for the corresponding certified maintenance personnel as reserved.
[0057] After the push is completed, a confirmation receipt from the maintenance management system is received. If no receipt is received within the timeout period, the work order sequence is resent. A receipt waiting timer is started after the push is completed, with a timeout threshold set to 30 seconds. This 30-second timeout threshold is based on the fact that the processing delay of the computerized maintenance management system's work order creation and resource reservation processes under normal working load typically does not exceed 15 seconds. Setting a 30-second timeout window provides sufficient time margin for system processing while avoiding excessive waiting that could affect the timeliness of maintenance response. If a confirmation receipt is received from the computerized maintenance management system within 30 seconds, containing a status code and a list of created work order numbers, the list of created work order numbers is recorded, and the resentment process is terminated. If no confirmation receipt is received after 30 seconds, the work order sequence push request is re-initiated, and the above push and confirmation receipt waiting process is repeated. If no confirmation receipt is received after three consecutive resentments, a push failure alarm is written to the maintenance log, and maintenance personnel are notified to manually intervene.
[0058] An intelligent equipment operation and maintenance system based on industrial big data includes a processor and a memory, which are connected. The memory stores programs, instructions, or code, while the processor executes these programs, instructions, or code to implement the aforementioned intelligent equipment operation and maintenance method based on industrial big data. In specific implementation, the processor uses a central processing unit (CPU) or a graphics processing unit (GPU), and the memory uses a non-volatile storage medium. The memory stores instructions for: acquiring vibration timing signals, temperature change sequences, and current waveform data during equipment operation; extracting spectral features, temperature change rate, and harmonic distortion; combining multi-dimensional feature vectors; calculating equipment health indicators using residuals; judging health indicators and triggering maintenance requests; parsing fault propagation chains to generate a list of suspected fault nodes; solving for the optimal maintenance plan using constraint logic programming; and outputting the optimal maintenance plan to the maintenance management system. The processor reads and executes these instructions from the memory to complete the entire intelligent equipment operation and maintenance process.
[0059] See Figure 5 In the graph, the horizontal axis represents time in minutes, ranging from 0 to 500, and the vertical axis represents the work order number, also ranging from 0 to 500. The curve in the graph represents the time sequence of work order tasks. The curve shows a gradual upward trend from the lower left to the upper right, indicating that the work order number gradually accumulates as time progresses, reflecting that the maintenance tasks unfold sequentially in time. Each work order is represented by a horizontal line segment at the corresponding time point, indicating the start and end times of the maintenance operation. The length of the line segment represents the estimated working hours of the work order. The dense line segments and relatively uniform length variation indicate that the working hours of the maintenance tasks are distributed continuously without significant interruptions. The sorting and distribution of this work order sequence conforms to the process in Example 5 of splitting the optimal maintenance plan into a time-ordered work order sequence. Each work order includes the maintenance object, maintenance operation code, required spare parts list, and estimated working hours. The continuity and time sequence of the work order tasks in the graph demonstrate that the optimal maintenance plan solved based on constraint logic programming reasonably arranges the timing of maintenance tasks, satisfies maintenance window and resource constraints, and achieves the orderly execution of maintenance tasks.
[0060] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.
Claims
1. A method for intelligent equipment operation and maintenance based on industrial big data, characterized in that, Includes the following steps: Collect vibration timing signals, temperature change sequences, and current waveform data during equipment operation; Fourier transform is performed on the vibration time sequence signal to extract spectral features, differential calculation is performed on the temperature change sequence to extract the temperature change rate, and harmonic distortion is extracted from the current waveform data. The spectral characteristics, temperature change rate, and harmonic distortion are combined into a multidimensional feature vector. Based on the residual calculation between the multidimensional feature vector and the physical model of equipment degradation, the equipment health index is obtained. When the health index falls below a preset threshold, a maintenance request is triggered. Based on the maintenance request, the fault propagation chain is analyzed from the equipment relationship graph to generate a list of suspected fault nodes; Based on the list of suspected fault nodes and maintenance history, the optimal maintenance solution is solved using constraint logic programming. The optimal repair solution is output to the maintenance management system.
2. The intelligent equipment operation and maintenance method based on industrial big data according to claim 1, characterized in that, The vibration timing signal, temperature change sequence, and current waveform data collected during the operation of the acquisition device specifically include: Accelerometers, thermocouples, and current transformers are installed at key measuring points of the equipment to collect data synchronously at a fixed sampling frequency. The collected raw data is segmented using a sliding window, and outliers exceeding three times the standard deviation within the window are removed. The data after removing outliers are aligned by timestamps to generate continuous vibration time-series signals, temperature change sequences, and current waveform data.
3. The intelligent equipment operation and maintenance method based on industrial big data according to claim 1, characterized in that, The extraction of spectral features by performing Fourier transform on the vibration time-series signal specifically includes: After applying a Hanning window function to the vibration time sequence signal, a fast Fourier transform is performed to obtain the amplitude spectrum; Extract the frequencies and amplitudes corresponding to the first five peak values from the amplitude spectrum to form a vibration spectrum feature vector; The vibration spectrum feature vector is normalized to a unit length and used as the spectrum feature.
4. The intelligent equipment operation and maintenance method based on industrial big data according to claim 1, characterized in that, The residual calculation based on the multidimensional feature vector and the device degradation physical model specifically includes: A physical model of equipment degradation is pre-established. This model takes a multi-dimensional feature vector as input and a health index as output. The model parameters are calibrated by finite element simulation. Calculate the Euclidean distance between the current multidimensional feature vector and the predicted vector output by the physical model under the same conditions; The Euclidean distance is processed by an exponentially weighted moving average to generate a smoothed equipment health index.
5. The intelligent equipment operation and maintenance method based on industrial big data according to claim 1, characterized in that, The step of parsing the fault propagation chain from the equipment relationship graph specifically includes: Read the equipment relationship graph, which uses equipment components as nodes and physical connections or process dependencies as directed edges; Starting from the equipment component that triggered the maintenance request, perform a reverse width-first search along the directed edges to obtain all reachable nodes as candidate fault nodes. Calculate the betweenness centrality of each candidate faulty node, and select nodes whose betweenness centrality exceeds a preset threshold to form a list of suspected faulty nodes.
6. The intelligent equipment operation and maintenance method based on industrial big data according to claim 5, characterized in that, The list of suspected faulty nodes, composed of nodes whose betweenness centrality exceeds a preset threshold, specifically includes: For each candidate faulty node, the ratio of the number of shortest paths passing through that node to the total number of shortest paths is calculated as the betweenness centrality. Sort the nodes by betweenness centrality from high to low, and take the top 30% of nodes as high-influence nodes; The list of suspected faulty nodes is generated by taking the union of the high-impact nodes and the nodes that have experienced faults in the maintenance history.
7. The intelligent equipment operation and maintenance method based on industrial big data according to claim 1, characterized in that, The method of using constraint logic programming to solve for the optimal maintenance solution specifically includes: The maintenance plan variable is defined as a binary decision variable, indicating whether to perform maintenance on each node in the list of suspected faulty nodes; Define a set of constraints, including a maximum total maintenance cost, a limit on the number of available maintenance personnel, and hard constraints that critical nodes must be maintained; With the optimization objective of minimizing the weighted sum of equipment downtime and maintenance costs, the constraint logic solver is invoked to search the feasible solution space. The maintenance scheme with the minimum objective function value is selected from the feasible solutions as the optimal maintenance scheme.
8. The intelligent equipment operation and maintenance method based on industrial big data according to claim 7, characterized in that, The set of defined constraints specifically includes: Obtain the current inventory quantity of repair parts from the enterprise resource planning system, and set a constraint that the quantity of repair consumables shall not exceed the inventory quantity; Obtain the schedule of certified personnel for each maintenance position from the human resources system, and set a constraint that no more than one maintenance task can be assigned to the same certified personnel within the same time period; The maintenance window time is set according to the production scheduling plan, and the execution time of the maintenance plan must be limited to the specified window.
9. The intelligent equipment operation and maintenance method based on industrial big data according to claim 1, characterized in that, The step of outputting the optimal repair solution to the maintenance management system specifically includes: The optimal repair plan is broken down into a sequence of work orders ordered by time. Each work order includes the repair object, repair operation code, required spare parts list, and estimated working hours. The work order sequence is pushed to the computerized maintenance management system through a standard interface, triggering the work order creation and resource reservation process. After the push is completed, a confirmation receipt is received from the maintenance management system. If no receipt is received within the time limit, the work order sequence is resent.
10. An intelligent equipment operation and maintenance system based on industrial big data, characterized in that: The device includes a processor and a memory, the memory and the processor being connected. The memory is used to store programs, instructions or code, and the processor is used to execute the programs, instructions or code in the memory to implement the intelligent operation and maintenance method for equipment based on industrial big data as described in any one of claims 1-9.
