Fault prediction method and system based on observable data

Abstract

The invention relates to the technical field of fault diagnosis, in particular to a fault prediction method and system based on observable data, and the method comprises the following steps: collecting real-time energy efficiency data, including power consumption, temperature and vibration data, of a gear and a bearing of mechanical equipment, calculating the deviation between the real-time energy efficiency data and a set energy efficiency reference, and calculating the energy efficiency of the gear and the bearing; and generating real-time energy efficiency deviation data. According to the method, the energy efficiency data of the key parts are monitored in real time, the abnormal state of the equipment is recognized in real time, the maintenance cost and downtime caused by fault diffusion are remarkably reduced, the deviation mode is analyzed through the random forest model, the abnormity judgment precision is improved, and the reliability of the equipment is improved. According to the method, fault propagation is analyzed by constructing the dependency graph and utilizing a graph theory method, the influence range and key nodes of potential faults are effectively predicted, the pertinence and efficiency of a maintenance strategy are greatly enhanced, and on the whole, the method improves the safety and reliability of a system, and meanwhile efficient operation of equipment is guaranteed.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and particularly to a fault prediction method and system based on observable data. Background Art

[0002] The technical field of fault diagnosis involves using automated and semi-automated systems, sensors, and algorithms to monitor, detect, analyze, and predict potential faults in equipment or systems. This technology is commonly applied in industrial automation, aviation, automotive, and other critical infrastructure sectors to ensure operational safety and improve efficiency. Fault diagnosis methods include real-time data monitoring, historical data analysis, pattern recognition, and machine learning techniques, aiming to identify abnormal behaviors early, prevent faults from occurring, thereby reducing downtime, lowering maintenance costs, and extending the lifespan of equipment.

[0003] Among them, the fault prediction method based on observable data refers to using the data collected from equipment or systems to predict estimated faults or problems. These data can include temperature, pressure, vibration levels, current, and other sensor outputs. The purpose of this method is that through continuous analysis of the data, future faults can be foreseen, enabling maintenance or component replacement in advance, effectively avoiding system downtime and serious equipment damage. The main goal of this technology is to improve predictive maintenance strategies, optimize resource utilization, and enhance overall operational efficiency and safety.

[0004] The existing technologies have deficiencies in real-time response and early fault identification. They usually rely on regular data monitoring and analysis, which cannot meet the requirements of rapidly changing equipment states and immediate handling of sudden faults. The lack of real-time data processing and precise fault propagation analysis limits the timeliness and accuracy of fault response, making it difficult to effectively predict and prevent system-level fault impacts. In complex mechanical systems, this limitation leads to failures in timely fault identification and handling, triggering a chain reaction, resulting in the obstruction of the entire system's functions, bringing high economic losses and a decline in production efficiency. Summary of the Invention

[0005] The purpose of the present invention is to solve the drawbacks existing in the prior art, and to propose a fault prediction method and system based on observable data.

[0006] To achieve the above purpose, the present invention adopts the following technical solution: A fault prediction method based on observable data, comprising the following steps: S1: Collect real-time energy efficiency data of the gears and bearings of mechanical equipment, including power consumption, temperature, and vibration data, calculate the deviation between the real-time energy efficiency data and the established energy efficiency benchmark, and generate real-time energy efficiency deviation data; S2: Based on the real-time energy efficiency deviation data, execute a random forest model to identify the deviation patterns of gear and bearing data, determine whether the deviation pattern is normal or abnormal, and obtain the abnormal pattern recognition result; S3: Use the abnormal pattern recognition result to construct a dependency graph between the motor and the hydraulic mechanism, analyze the stability of the dependency relationship and the potential for fault propagation, predict the propagation path and influence range of the fault through graph theory methods, and generate a fault impact prediction network; S4: According to the fault impact prediction network, analyze the key nodes in the motor and the hydraulic system, calculate the centrality of each node, and identify the nodes at risk due to the fault impact to obtain the key node fault risk analysis result.

[0007] As a further solution of the present invention, the steps for obtaining the real-time energy efficiency data are specifically as follows: S111: Read the power consumption, temperature, and vibration levels of the gears and bearings of the mechanical equipment from the sensors to generate the pre-processed real-time energy efficiency data; S112: Pre-process the pre-processed real-time energy efficiency data to eliminate outliers and fill in missing values, and use the data standardization formula:

[0008] Obtain the cleaned real-time energy efficiency data; Wherein, is the cleaned real-time energy efficiency data, represents a single data point, is the mean of the data, is the standard deviation; S113: From the cleaned real-time energy efficiency data, through the weighted sum formula:

[0009] Calculate and generate the key performance indicators; Wherein, is the key performance indicator, is the power data point, is the temperature data point, is the vibration data point, 、 、 are the weight coefficients of power, temperature, and vibration data respectively; S114: Use the key performance indicators, through the formula:

[0010] Calculate the real-time energy efficiency comprehensive index to obtain the real-time energy efficiency data; Wherein, is the comprehensive index of the real-time energy efficiency, Indicates the weight of the differential performance index, is the adjustment factor to balance the contributions among multiple indicators.

[0011] As a further solution of the present invention, the steps of the real-time energy efficiency deviation data are specifically as follows: S121: Compare the real-time energy efficiency data with the preset energy efficiency benchmark, and generate a benchmark comparison result by comparing the real-time data and the benchmark data of each measurement point; S122: For the benchmark comparison result, use the formula:

[0012] Calculate the deviation ratio of each data point to generate deviation data; Among them, is the deviation ratio of the th data point, represents the real-time energy efficiency data point, is the corresponding energy efficiency benchmark value, is a small positive number to avoid the denominator being zero; S123: Use the deviation data for statistical analysis, and use the formula:

[0013]

[0014] Calculate the average deviation and the standard deviation of the deviation to obtain the statistical deviation analysis result; Among them, is the total number of data points, Avg_ is the average value of the deviation data, is the standard deviation of the deviation data; S124: Organize the statistical deviation analysis result to generate an evaluation result, forming the real-time energy efficiency deviation data.

[0015] As a further solution of the present invention, the steps of obtaining the abnormal mode recognition result are specifically as follows: S211: Calculate the standardized deviation value of each monitoring point from the real-time energy efficiency deviation data, using the formula:

[0016] Generate the standardized deviation data; Among them, is the standardized deviation, is the deviation of a single data point, is the parameter to adjust the influence of standardization; S212: Apply threshold analysis to the standardized deviation data, set a threshold , and use the formula:

[0017] to obtain the threshold analysis result; where is the result of each data point after threshold analysis, is a parameter that controls the steepness of the curve; S213: Analyze the threshold analysis result, and use the weighted clustering analysis formula:

[0018] to generate a weighted abnormal pattern recognition result; where is the weighted result of the clustering analysis, is the weight of each data point; S214: Synthesize the weighted abnormal pattern recognition result, and use the majority decision rule formula:

[0019] to determine the key abnormal pattern and obtain the abnormal pattern recognition result.

[0020] As a further solution of the present invention, the method for analyzing the stability of the dependency relationship and the potential of fault propagation is specifically: S311: Use the data obtained from the abnormal pattern recognition result to calculate the dependency strength between devices, and use the formula:

[0021] to generate a dependency strength matrix; where is the dependency strength matrix, , are the abnormal state indicators of devices and , is the attenuation factor, is the coefficient for adjusting the distance, , are the physical distances between devices; S312: Based on the dependency strength matrix, perform centrality analysis to evaluate the influence and stability of each node, and use the formula:

[0022] to calculate the centrality of each node and generate a node centrality analysis result; where is the node Centrality, is the total number of nodes, , are the load capacities of nodes and respectively; is the influence coefficient of load difference on dependence; S313: Using the node centrality analysis result, evaluate the fault propagation potential of multiple nodes, using the formula:

[0023] Obtain the fault propagation potential evaluation result; wherein, is the fault propagation potential index of node ; is the centrality value of node ; is the normalization coefficient for balancing the fault propagation probability.

[0024] As a further solution of the present invention, the obtaining steps of the fault impact prediction network are specifically as follows: S321: Based on the stability of the dependency relationship and the fault propagation potential data, use graph theory analysis to identify the betweenness centrality of each node, using the formula:

[0025] Generate the node betweenness centrality analysis result; wherein, is the betweenness centrality of node ; is the total number of the shortest paths between nodes and ; is the number of paths passing through node ; S322: Using the node betweenness centrality analysis result, determine the critical fault propagation path, using the formula:

[0026] Obtain the critical fault propagation path; wherein, is the importance score of the fault propagation path, is the load of node ; is the total load of nodes in the network; S323: Based on the critical fault propagation path, evaluate the factors that the faults on the path affect the entire network, using the formula:

[0027] Obtain the fault impact scope; Among them, is the predicted value of the fault impact scope, is the node 's fault probability; Integrate the node betweenness centrality analysis results, the critical fault propagation paths, and the fault impact scope to generate a fault impact prediction network.

[0028] As a further solution of the present invention, the step of calculating the centrality of each node is specifically as follows: S411: Obtain the data of the fault impact prediction network from the maintenance system, including nodes, connection types, and strengths, to obtain the network structure data; S412: Analyze the network structure data and use graph theory methods:

[0029] Determine the critical nodes in the motor and hydraulic pressure systems and calculate the weighted degree centrality of each node; Among them, is the original node degree centrality, represents the existence of connections between nodes, represents the weight of the connection; S413: For the nodes in the list of critical nodes, adopt the formula:

[0030] Calculate the centrality of each node; Among them, is the adjusted node centrality, is the sum of the original degree centralities of the nodes in the network, is the node 's fault occurrence frequency, is the maximum fault frequency, is the total number of nodes in the network.

[0031] As a further solution of the present invention, the step of obtaining the analysis result of the critical node fault risk is specifically as follows: S421: According to the centrality of the node, apply the normalized risk factor formula:

[0032] Generate node risk factors; Among them, represents the risk factor of the th node, is the total number of nodes; S422: Based on multiple said node risk factors Average value , standard deviation , substitute into system sensitivity and environmental factor , adopt the formula:

[0033] Calculate to obtain the risk threshold ; Among them, is the risk threshold, and 1.5 is the weight coefficient; S423: Then apply and comparison, obtain a logical array:

[0034] Identify the nodes at risk, and summarize the risk node logical array to obtain the key node failure risk analysis result; Among them, is the indicator function.

[0035] A fault prediction system based on observable data, the fault prediction system based on observable data is used to execute the above-mentioned fault prediction method based on observable data, and the system includes: The data acquisition module monitors the power consumption, temperature, and vibration data of the mechanical equipment gears and bearings in real time, calculates the deviation between the data and the established energy efficiency benchmark, and generates real-time energy efficiency deviation data; The deviation analysis module analyzes the real-time energy efficiency deviation data, identifies the mutual relationship and change trend between the data, extracts and summarizes the change trend, and forms deviation pattern data; The abnormal pattern recognition module screens out prominent abnormal patterns from the deviation pattern data, compares the abnormal patterns with the normal operation patterns, identifies the data points that do not meet the normal operation standards, and generates abnormal pattern recognition results; The dependency graph construction module uses the abnormal pattern recognition result to construct the mutual dependency graph between multiple components in the mechanical system, including the motor and the hydraulic mechanism, analyzes the connectivity between multiple nodes and its impact on the system stability, and creates a fault impact prediction network; The key node analysis module performs centrality analysis on the nodes in the network according to the fault impact prediction network, screens out the key nodes affected by fault propagation, and obtains the key node failure risk analysis result.

[0036] Compared with the prior art, the advantages and positive effects of the present invention are: In the present invention, by monitoring the energy efficiency data of key components in real time, the instant identification of abnormal states of the equipment is realized, thereby significantly reducing the maintenance costs and downtime that may be caused by the spread of faults. By using the random forest model to analyze deviation patterns, the accuracy of abnormal determination is improved. By constructing a dependency graph and using graph theory methods to analyze fault propagation, the influence range and key nodes of potential faults are effectively predicted, greatly enhancing the pertinence and efficiency of maintenance strategies. Overall, the method improves the security and reliability of the system, and at the same time ensures the efficient operation of the equipment. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 is a schematic diagram of the work flow of the present invention; Figure 2 is a flow chart of the steps for obtaining real-time energy efficiency data of the present invention; Figure 3 is a flow chart of the steps for real-time energy efficiency deviation data of the present invention; Figure 4 is a flow chart of the steps for obtaining the recognition results of abnormal patterns of the present invention; Figure 5 is a flow chart of the method for analyzing the stability of dependency relationships and the potential of fault propagation of the present invention; Figure 6 is a flow chart of the steps for calculating the centrality of each node of the present invention; Figure 7 is a flow chart of the steps for obtaining the fault impact prediction network of the present invention; Figure 8 is a flow chart of the steps for obtaining the analysis results of the fault risk of key nodes of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0038] In order to make the objectives, technical solutions and advantages of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0039] In the description of the present invention, it should be understood that the orientation or positional relationships indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. are based on the orientation or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore should not be construed as limiting the present invention. In addition, in the description of the present invention, "a plurality of" means two or more unless otherwise specifically defined.

[0040] Embodiment 1 Please refer to Figure 1 , the present invention provides a technical solution: a fault prediction method based on observable data, comprising the following steps: S1: Collect real-time energy efficiency data of the gears and bearings of mechanical equipment, including power consumption, temperature, and vibration data, calculate the deviation between the real-time energy efficiency data and the established energy efficiency benchmark, and generate real-time energy efficiency deviation data; S2: Based on the real-time energy efficiency deviation data, execute a random forest model to identify the deviation patterns of the gear and bearing data, determine whether the deviation pattern is normal or abnormal, and obtain the abnormal pattern recognition result; S3: Use the abnormal pattern recognition result to construct a dependency graph between the motor and the hydraulic mechanism, analyze the stability of the dependency relationship and the potential for fault propagation, predict the propagation path and scope of influence of the fault through graph theory methods, and generate a fault impact prediction network; S4: According to the fault impact prediction network, analyze the key nodes in the motor and hydraulic system, calculate the centrality of each node, and identify the nodes at risk due to the fault impact to obtain the key node fault risk analysis result.

[0041] The real-time energy efficiency deviation data includes the energy consumption deviation value, the temperature difference value, and the vibration frequency deviation. The abnormal pattern recognition result is specifically the normal state index, the abnormal state index, and the state deviation type. The fault impact prediction network includes the relationship connectivity graph, the stability score, and the fault propagation path. The key node fault risk analysis includes node centrality, risk nodes, and impact assessment.

[0042] Please refer to Figure 2 , the specific steps for obtaining the real-time energy efficiency data are as follows: S111: Read the power consumption, temperature, and vibration levels of the gears and bearings of the mechanical equipment from the sensors to generate the real-time energy efficiency data before processing; S112: Preprocess the real-time energy efficiency data before processing to eliminate outliers and fill in missing values, and use the data standardization formula:

[0043] to obtain the cleaned real-time energy efficiency data; wherein, is the cleaned real-time energy efficiency data, represents a single data point, is the mean value of the data, is the standard deviation; S113: From the cleaned real-time energy efficiency data, through the weighted sum formula:

[0044] calculate and generate the key performance indicators; wherein, is a key performance indicator, is a power data point, is a temperature data point, is a vibration data point, , , are the weight coefficients of power, temperature, and vibration data respectively; S114: Using the key performance indicator, through the formula:

[0045] calculate the real-time comprehensive energy efficiency index to obtain real-time energy efficiency data; Among them, is the comprehensive index of real-time energy efficiency, represents the weight of the differential performance index, is the adjustment factor to balance the contributions among multiple indicators.

[0046] Formula:

[0047] Explanation: : A single raw data point.

[0048] : The mean of the data set.

[0049] : The standard deviation of the data set.

[0050] Derivation process and example: Assume that the monitored raw temperature data point is , and from past data, we know the average temperature recorded by this sensor and the standard deviation .

[0051] Substitute into the formula:

[0052] This means that this temperature data point is one standard deviation higher than the average temperature.

[0053] Formula:

[0054] Explanation: : Power data point, assume watts.

[0055] : Temperature data point, given in the previous example .

[0056] : Vibration data points, assuming , representing acceleration.

[0057] : Weighting factor, set according to the system design priority. Assuming .

[0058] Derivation process and example:

[0059]

[0060] This formula combines the effects of power, temperature, and vibration into a single performance metric.

[0061] Formula:

[0062] Explanation: : Weights of different performance metrics, assuming .

[0063] : Adjustment factor, set to to simplify the calculation.

[0064] : Number of performance metrics, assuming there is one metric .

[0065] Derivation process and example:

[0066] This formula calculates a comprehensive real-time energy efficiency metric.

[0067] Please refer to Figure 3 , the steps for real-time energy efficiency deviation data are as follows: S121: Compare the real-time energy efficiency data with the preset energy efficiency benchmark. By comparing the real-time data and the benchmark data for each measurement point, generate a benchmark comparison result; S122: For the benchmark comparison result, use the formula:

[0068] Calculate the deviation ratio for each data point to generate deviation data; Among them, is the deviation ratio of the th data point, represents the real-time energy efficiency data point, is the corresponding energy efficiency benchmark value, is a small positive number to avoid a zero denominator; S123: Perform statistical analysis using the deviation data, adopting the formula:

[0069]

[0070] Calculate the average deviation and the standard deviation of the deviation Obtain the statistical deviation analysis result; Among them, is the total number of data points, Avg_ is the average value of the deviation data, is the standard deviation of the deviation data; S124: Organize the statistical deviation analysis result to generate an evaluation result, forming real-time energy efficiency deviation data.

[0071] Formula:

[0072] Explanation: : The real-time energy efficiency value of the th data point.

[0073] : The energy efficiency benchmark value of the th data point.

[0074] : A small positive number used to avoid division by zero, set to 0.01.

[0075] Calculation process: Obtain the real-time energy efficiency and the benchmark energy efficiency value: Assume the real-time energy efficiency value watts.

[0076] Assume the benchmark energy efficiency value watts.

[0077] Calculate the deviation: Deviation watts.

[0078] Apply the formula:

[0079] Formula:

[0080] Calculation process: Assume the deviations of five data points: .

[0081] Apply the formula: Avg 。

[0082] Formula:

[0083] Calculation process: Use the above deviation value and average deviation:

[0084]

[0085]

[0086]

[0087] Please refer to Figure 4 , and the specific steps for obtaining the abnormal pattern recognition result are as follows: S211: Calculate the standardized deviation value of each monitoring point from the real-time energy efficiency deviation data, using the formula:

[0088] Generate standardized deviation data; Among them, is the standardized deviation, is the deviation of a single data point, is the parameter for adjusting the influence of standardization; S212: Apply threshold analysis to the standardized deviation data, set the threshold , and use the formula:

[0089] Obtain the threshold analysis result; Among them, is the result of each data point after threshold analysis, is the parameter for controlling the steepness of the curve; S213: Analyze the threshold analysis result, using the weighted clustering analysis formula:

[0090] Generate the weighted abnormal pattern recognition result; Among them, is the weighted result of the clustering analysis, is the weight of each data point; S214: Synthesize the weighted abnormal pattern recognition result, using the majority decision rule formula:

[0091] Identify the key abnormal patterns to obtain the abnormal pattern recognition results; Among them, is the abnormal pattern recognition result.

[0092] Formula:

[0093] Parameter explanation and derivation: : The deviation of a single monitoring point. Assume there is a monitoring value , considering .

[0094] and These are the minimum and maximum values in the deviation array. For a given array , the minimum value , the maximum value .

[0095] : A parameter that adjusts the influence of logarithmic normalization, taking as a reasonable example value.

[0096] Calculation process: First, calculate the normalization part:

[0097] Calculate the logarithmic part (increasing the influence of ):

[0098] Combine the two parts to obtain :

[0099] Calculation of the threshold analysis result Formula:

[0100]

[0101] Calculate the logarithmic part (increasing the influence of ):

[0102] Combine the two parts to obtain :

[0103] Formula:

[0104] Parameter Explanation and Deduction: : Controls the steepness of the function curve, set to , based on the sensitivity of a typical logic function.

[0105] : Threshold, set to , assuming this is the boundary between abnormal and normal.

[0106] Calculation Process: Use :

[0107] Formula:

[0108] Parameter Explanation and Deduction: : Weight parameter, assuming corresponds to the importance of different monitoring points.

[0109] Calculation Process: For a single data point, use and :

[0110] Formula:

[0111] Consider a simple example, the value of which is 0.718, so .

[0112] Please refer to Figure 5 for the method of analyzing the stability of dependency relationships and the potential of fault propagation, specifically: S311: Use the data obtained from the abnormal pattern recognition results to calculate the dependency strength between devices, using the formula:

[0113] Generate a dependency strength matrix; where, is the dependency strength matrix, , are the abnormal state indicators of devices and , is the attenuation factor, is the coefficient for distance adjustment, , are the physical distances between devices; S312: Based on the dependency strength matrix, perform centrality analysis to evaluate the influence and stability of each node, using the formula:

[0114] Calculate the centrality of each node to generate the node centrality analysis result; Among them, is the centrality of node , is the total number of nodes, , are the load capacities of nodes and respectively, is the influence coefficient of load difference on dependency; S313: Use the node centrality analysis result to evaluate the fault propagation potential of multiple nodes, using the formula:

[0115] Obtain the fault propagation potential evaluation result; Among them, is the fault propagation potential index of node , is the centrality value of node , is the normalization coefficient used to balance the fault propagation probability.

[0116] Formula:

[0117] Parameter explanation: : Abnormal state indicators of devices and . Assume (mild abnormality) and (basically normal).

[0118] : Coefficient for controlling the attenuation of dependency strength. Set to , which is a commonly used value to enhance the influence of abnormal state indicators.

[0119] : Coefficient for adjusting the influence of the physical distance between devices. Set to .

[0120] Physical distance between devices. Assume meters and meters.

[0121] Calculation process: Calculate the differential effect of the abnormal state index: . Calculate the sum of the distance adjustment factors: . Calculate the strength of the dependency:

[0122] (Since is close to which is close to 1, indicating a very strong dependency) Formula:

[0123] Parameter explanation: : The strength of the dependency calculated in the previous step.

[0124] : The total number of nodes in the graph. Assume there are nodes.

[0125] : The load capacity of the node. Assume units, units.

[0126] : The coefficient of the influence of the load difference on the dependency, set to .

[0127] Calculation process: For each , calculate the load difference effect: . Calculate the centrality score:

[0128] (This indicates that the centrality of node is relatively low because the load difference is large) Formula:

[0129] Parameter explanation: : The centrality of the node calculated in the previous step.

[0130] : The total number of nodes.

[0131] : The coefficient for normalizing the probability of fault propagation, set to .

[0132] Calculation process: Calculate the sum of the indices of all node centralities:

[0133] Calculate the fault propagation potential:

[0134] (This indicates that the fault propagation potential of node is relatively low.) Please refer to Figure 6 , and the specific steps for obtaining the fault impact prediction network are as follows: S321: Based on the stability and fault propagation potential data of dependencies, use graph theory analysis to identify the betweenness centrality of each node, and adopt the formula:

[0135] Generate the betweenness centrality analysis results of the nodes; Among them, is the betweenness centrality of node , is the total number of shortest paths between node and , is the number of paths passing through node ; S322: Utilize the betweenness centrality analysis results of the nodes to determine the critical fault propagation paths, and adopt the formula:

[0136] Obtain the critical fault propagation paths; Among them, is the importance score of the fault propagation path, is the load of node , is the total load of the nodes in the network; S323: Based on the critical fault propagation paths, evaluate the factors that the faults on the paths affect the entire network, and adopt the formula:

[0137] Obtain the fault impact range; Among them, is the predicted value of the fault impact range, is the fault probability of node ; Integrate the betweenness centrality analysis results of the nodes, the critical fault propagation paths, and the fault impact range to generate the fault impact prediction network.

[0138] Formula:

[0139] : Betweenness centrality of a node represents the frequency of occurrence of the node in the shortest paths considering all nodes .

[0140] : Total number of all shortest paths between node and .

[0141] : Number of shortest paths passing through node .

[0142] Calculation process: Assume a small network with the number of nodes , calculate the betweenness centrality of node 2 . Assume the number of shortest paths from node 1 to node 3 passing through node 2 , and the total number of shortest paths from all node 1 to node 3 . Similarly, calculate other paths

[0143] Calculate :

[0144]

[0145] Formula:

[0146] Importance score of the fault propagation path

[0147] : Betweenness centrality of node in the path

[0148] : Load of node

[0149] : Total load of all nodes in the network

[0150] Calculation process: Assume the path includes node , where .

[0151]

[0152]

[0153] ​ :

[0154] : Predicted value of the fault influence range.

[0155] : Node 's fault probability.

[0156] Calculation process: Assume .

[0157]

[0158] Please refer to Figure 7 , the steps to calculate the centrality of each node are specifically as follows: S411: Obtain the data of the fault influence prediction network from the maintenance system, including nodes, connection types and strengths, to get the network structure data; S412: Analyze the network structure data and use graph theory methods:

[0159] Determine the key nodes in the motor and hydraulic pressure systems, and calculate the weighted degree centrality of each node; Among them, is the original node degree centrality, represents the existence of connections between nodes, represents the weight of the connection; S413: For the nodes in the list of key nodes, use the formula:

[0160] Calculate the centrality of each node; Among them, is the adjusted node centrality, is the sum of the original degree centralities of the nodes in the network, is the node 's fault occurrence frequency, is the maximum fault frequency, is the total number of nodes in the network.

[0161] Formula:

[0162] Parameter explanation: Node 's weighted degree centrality.

[0163] : Represents the node Whether there is a direct connection between the node and the node usually takes a value of 0 or 1.

[0164] : Connection and The weight between them can be determined according to the importance of the connection, traffic, transmission rate, or other system-specific metrics.

[0165] Derivation process and example: Suppose there is a network with three nodes, and the connection situation and weights between the nodes are as follows: The connection weight between node 1 and node 2 is 0.5.

[0166] The connection weight between node 1 and node 3 is 0.3.

[0167] There is no direct connection between node 2 and node 3.

[0168] Therefore, the connection matrix and the weight matrix are as follows:

[0169]

[0170] For the calculation of the weighted degree centrality of node 1 :

[0171] Formula:

[0172] Parameter explanation: Adjusted node centrality.

[0173] : The sum of the original degree centralities of all nodes.

[0174] Node The failure occurrence frequency.

[0175] The maximum failure frequency in the network.

[0176] Derivation process and example: Continue to use the above network: Suppose (failure frequency), and suppose .

[0177] and can also be calculated similarly. Assume .

[0178] Formula:

[0179] Parameter Explanation: Adjusted node centrality.

[0180] : The sum of the original degree centralities of all nodes.

[0181] Node 's failure occurrence frequency.

[0182] The maximum failure frequency in the network.

[0183] Derivation Process and Numerical Example: Continue to use the above network: Assume (failure frequency), and assume .

[0184] and can also be calculated similarly. Assume .

[0185] Please refer to Figure 8 , and the specific steps for obtaining the analysis results of the critical node failure risk are as follows: S421: According to the centrality of the node , apply the normalized risk factor formula:

[0186] Generate the node risk factor; Among them, represents the risk factor of the th node, is the total number of nodes; S422: Based on the average value of multiple node risk factors , standard deviation , substitute into the system sensitivity and environmental factor , and adopt the formula:

[0187] Calculate to obtain the risk threshold ; Among them, is the risk threshold, and 1.5 is the weight coefficient; S423: Then apply with to compare and obtain a logical array:

[0188] Identify the nodes at risk and summarize the logical array of risk nodes to obtain the analysis result of the critical node failure risk; Among them, is the indicator function.

[0189] Formula:

[0190] Here represents the risk factor of the th node, is the centrality of node , is the total number of nodes.

[0191] Parameter explanation: : The centrality of the th node, which can be obtained through network analysis methods such as degree centrality and betweenness centrality.

[0192] : The sum of the centralities of all nodes, ensuring normalization.

[0193] Calculation process: Assume that there are three nodes in the network, and their centralities are 5, 3, and 2 respectively.

[0194] Total centrality

[0195] For the first node:

[0196] Formula:

[0197] : The average value of.

[0198] : The standard deviation of.

[0199] : The system sensitivity parameter, which reflects the reaction of the system to external changes.

[0200] : Environmental factor, reflecting the impact of the external environment.

[0201] Logical array:

[0202] : Indicator function, indicating whether it exceeds the above.

[0203] Calculation process: Assume ; Average value

[0204] Standard deviation

[0205] Assume (Parameter selection is based on the assumed system sensitivity and the range of average environmental factors); Risk threshold

[0206] , so (Not marked as a risk node) A fault prediction system based on observable data, and the fault prediction system based on observable data is used to execute the above-mentioned fault prediction method based on observable data. The system includes: The data acquisition module monitors the power consumption, temperature, and vibration data of the gears and bearings of the mechanical equipment in real time, calculates the deviation between the data and the established energy efficiency benchmark, and generates real-time energy efficiency deviation data; The deviation analysis module analyzes the real-time energy efficiency deviation data, identifies the mutual relationships and change trends between the data, extracts and generalizes the change trends, and forms deviation pattern data; The abnormal pattern recognition module screens out prominent abnormal patterns from the deviation pattern data, compares the abnormal patterns with the normal operation patterns, identifies the data points that do not meet the normal operation standards, and generates abnormal pattern recognition results; The dependency graph construction module uses the abnormal pattern recognition results to construct the mutual dependency graph between multiple components in the mechanical system, including the motor and the hydraulic mechanism, analyzes the connectivity between multiple nodes and its impact on the system stability, and creates a fault impact prediction network; The critical node analysis module performs centrality analysis on the nodes in the fault impact prediction network, screens out the critical nodes affected by the fault propagation, and obtains the critical node fault risk analysis results.

[0207] The above are only the preferred embodiments of the present invention and do not limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A fault prediction method based on observable data, characterized in that, It includes the following steps: Collect real-time energy efficiency data of the gears and bearings of mechanical equipment, including power consumption, temperature, and vibration data, calculate the deviation between the real-time energy efficiency data and the established energy efficiency benchmark, and generate real-time energy efficiency deviation data; Based on the real-time energy efficiency deviation data, execute a random forest model to identify the deviation patterns of gear and bearing data, determine whether the deviation pattern is normal or abnormal, and obtain the abnormal pattern recognition result; Use the abnormal pattern recognition result to construct a dependency graph between the motor and the hydraulic mechanism, analyze the stability of the dependency relationship and the potential for fault propagation, predict the propagation path and influence range of the fault through graph theory methods, and generate a fault impact prediction network; According to the fault impact prediction network, analyze the key nodes in the motor and hydraulic system, calculate the centrality of each node, and identify the nodes at risk due to the fault impact to obtain the key node fault risk analysis result.

2. The fault prediction method based on observable data according to claim 1, wherein The specific steps for obtaining the real-time energy efficiency data are as follows: Read the power consumption, temperature, and vibration levels of the gears and bearings of the mechanical equipment from the sensors to generate pre-processed real-time energy efficiency data; Pre-process the pre-processed real-time energy efficiency data to eliminate outliers and fill in missing values, using the data standardization formula: ； Obtain the cleaned real-time energy efficiency data; Among them, is the real-time energy efficiency data after cleaning, represents a single data point, is the mean value of the data, is the standard deviation; From the cleaned real-time energy efficiency data, through the weight summation formula: ； Calculate and generate key performance indicators; Among them, is a key performance indicator, is a power data point, is a temperature data point, is a vibration data point, , , are the weight coefficients of power, temperature, and vibration data, respectively; Use the key performance indicators, through the formula: ； Calculate the real-time energy efficiency comprehensive index to obtain the real-time energy efficiency data; Among them, is the comprehensive index of real-time energy efficiency, represents the weight of the differential performance index, is the adjustment factor to balance the contributions among multiple indicators.

3. The fault prediction method based on observable data according to claim 1, wherein The steps for the real-time energy efficiency deviation data are as follows: Compare the real-time energy efficiency data with the preset energy efficiency benchmark, generate a benchmark comparison result by comparing the real-time data and the benchmark data at each measurement point; For the benchmark comparison result, use the formula: ； Calculate the deviation ratio of each data point to generate deviation data; Among them, is the deviation ratio of the th data point, represents the real-time energy efficiency data point, is the corresponding energy efficiency benchmark value, is a small positive number to avoid a zero denominator; Use the deviation data for statistical analysis, using the formula: ； ； Calculate the average deviation and the standard deviation of the deviation Obtain the statistical deviation analysis result; Among them, is the total number of data points, Avg_ is the average value of the deviation data, is the standard deviation of the deviation data; Organize the statistical deviation analysis results to generate an evaluation result, forming real-time energy efficiency deviation data.

4. The fault prediction method based on observable data according to claim 1, wherein The specific steps for obtaining the abnormal pattern recognition result are as follows: Calculate the standardized deviation value of each monitoring point from the real-time energy efficiency deviation data, using the formula: ； Generate standardized deviation data; wherein, is the deviation after standardization, is the deviation of a single data point, is a parameter for adjusting the influence of standardization; Apply threshold analysis to the standardized deviation data and set a threshold , and use the formula: ； Obtain the threshold analysis result; Among them, is the result of each data point after threshold analysis, is a parameter that controls the steepness of the curve; Analyze the threshold analysis result, using the weighted clustering analysis formula based on weights: ； Generate a weighted abnormal pattern recognition result; Among them, is the weighted result of cluster analysis, is the weight of each data point; Integrate the weighted abnormal pattern recognition results, using the majority decision rule formula: ； Determine the key abnormal pattern to obtain the abnormal pattern recognition result.

5. The fault prediction method based on observable data according to claim 1, wherein The method for analyzing the stability of the dependency relationship and the potential for fault propagation is as follows: Use the data obtained from the abnormal pattern recognition result to calculate the dependency strength between devices, using the formula: ； Generate a dependency strength matrix; Among them, is the dependence intensity matrix, , are the abnormal state indicators of devices and , is the attenuation factor, is the coefficient for distance adjustment, , are the physical distances between devices; Based on the dependency strength matrix, perform centrality analysis to evaluate the influence and stability of each node, using the formula: ； Calculate the centrality of each node to generate a node centrality analysis result; Among them, is the centrality of the node , is the total number of nodes, , are the load capacities of nodes and respectively, is the influence coefficient of the load difference on the dependence; Use the node centrality analysis result to evaluate the potential for fault propagation of multiple nodes, using the formula: ； Obtain the fault propagation potential evaluation result; Among them, is the fault propagation potential index of the node , is the centrality value of the node , is the normalization coefficient used to balance the fault propagation probability.

6. The fault prediction method based on observable data according to claim 1, wherein The specific steps for obtaining the fault impact prediction network are as follows: Based on the stability and fault propagation potential data of the said dependencies, use graph theory analysis to identify the betweenness centrality of each node, adopting the formula: ； Generate the betweenness centrality analysis results of the nodes; Among them, is the betweenness centrality of node , is the total number of shortest paths between node and , is the number of paths passing through node . Utilize the betweenness centrality analysis results of the nodes to determine the critical fault propagation paths, adopting the formula: ； Obtain the critical fault propagation paths; Among them, is the importance score of the fault propagation path, is the node load, is the total load of the nodes in the network; Based on the critical fault propagation paths, evaluate the factors for the faults on the paths to affect the entire network, adopting the formula: ； Obtain the fault impact scope; Among them, is the predicted value of the fault influence range, is the node fault probability; Integrate the betweenness centrality analysis results of the nodes, the critical fault propagation paths and the fault impact scope to generate a fault impact prediction network.

7. The fault prediction method based on observable data according to claim 1, wherein The steps of calculating the centrality of each node are specifically as follows: Obtain the data of the fault impact prediction network from the maintenance system, including nodes, connection types and strengths, to get the network structure data; Analyze the network structure data, using graph theory methods: ； Determine the critical nodes in the motor and the hydraulic pressure system, and calculate the weighted degree centrality of each node; Among them, is the original node degree centrality, represents the existence of connections between nodes, represents the weight of the connection; For the nodes in the list of the critical nodes, adopt the formula: ； Calculate the centrality of each node; wherein, is the adjusted node centrality, is the sum of the original degree centralities of the nodes in the network, is the node fault occurrence frequency, is the maximum fault frequency, is the total number of nodes in the network.

8. The fault prediction method based on observable data according to claim 1, characterized in that, The steps of obtaining the critical node fault risk analysis results are specifically as follows: According to the centrality of the said node , apply the normalized risk factor formula: ； Generate node risk factors; Among them, represents the risk factor of the th node, where is the total number of nodes; Based on the average values of multiple said node risk factors , standard deviation , substitute into the system sensitivity and environmental factors , and use the formula: ​ ； The calculated risk threshold ; Among them, is the risk threshold, and 1.5 is the weight coefficient; Apply again and comparison to obtain a logical array: ； Identify the nodes at risk and summarize the logical array of risk nodes to obtain the analysis result of the critical node failure risk; wherein, is an indicator function.

9. A fault prediction system based on observable data, characterized in that, According to the fault prediction method based on observable data according to any one of claims 1-8, the system includes: The data acquisition module monitors the power consumption, temperature and vibration data of the gears and bearings of the mechanical equipment in real time, calculates the deviation between the data and the established energy efficiency benchmark, and generates real-time energy efficiency deviation data; The deviation analysis module analyzes the real-time energy efficiency deviation data, identifies the mutual relationships and change trends between the data, extracts and generalizes the change trends, and forms deviation pattern data; The abnormal pattern recognition module screens out the prominent abnormal patterns from the deviation pattern data, compares the abnormal patterns with the normal operation patterns, identifies the data points that do not meet the normal operation standards, and generates abnormal pattern recognition results; The dependency graph construction module uses the abnormal pattern recognition results to construct the mutual dependency graph of multiple components in the mechanical system, including the motor and the hydraulic pressure mechanism, analyzes the connectivity between multiple nodes and its impact on the system stability, and creates a fault impact prediction network; The critical node analysis module conducts centrality analysis on the nodes in the network according to the fault impact prediction network, screens out the critical nodes affected by fault propagation, and obtains the critical node fault risk analysis results.

Images