A method and system for intelligent diagnosis and processing of equipment fault data

Abstract

This invention relates to the field of data processing, specifically to an intelligent diagnostic processing method and system for equipment fault data. It includes: acquiring historical equipment fault events; constructing a fault decision tree based on these events, the construction process of which involves: extracting features from time-series data in each window to form feature vectors; clustering all feature vectors to convert each historical equipment fault event into a symbolic sequence; grouping symbolic sequences with the same cause label into the same group to obtain grouping results; using the normal state of all historical equipment fault events as the root node, iteratively generating nodes at each level of each decision branch; converting real-time equipment fault data into real-time symbolic sequences, matching the paths corresponding to the real-time symbolic sequences to obtain the target path, and calculating the probability of equipment fault risk based on the target path. This invention can improve the timeliness of responding to equipment fault events.

Description

Technical Field

[0001] This invention relates to the field of data processing, and more specifically to an intelligent diagnostic processing method and system for equipment fault data. Background Technology

[0002] In current industrial scenarios, equipment failure handling solutions are typically obtained after a failure has occurred and been detected, relying on expert experience for judgment and decision-making. This "post-event response" model keeps the handling process in a reactive state, making it difficult to intervene effectively in the early stages of a failure.

[0003] The key reason for this problem is that existing detection methods mostly use threshold alarm mechanisms, which only trigger a response when equipment parameters significantly exceed safe limits. By this time, the fault has already developed to a relatively serious stage, causing not only equipment damage or production interruption but also severely limiting the options for subsequent handling solutions.

[0004] More importantly, relying on human experience to formulate solutions lacks systematic mining and quantitative analysis of historical fault data. This makes solution selection subjective and lagging, making it difficult to provide timely and effective intervention measures for specific fault types and their evolution patterns. Summary of the Invention

[0005] This invention provides an intelligent diagnostic processing method and system for equipment fault data to solve existing problems.

[0006] The intelligent diagnostic processing method for equipment fault data of the present invention adopts the following technical solution:

[0007] One embodiment of the present invention provides an intelligent diagnostic processing method for equipment fault data, the method comprising the following steps:

[0008] Acquire historical equipment failure events, wherein the historical equipment failure events include at least two types of time-series data, and each historical equipment failure event corresponds to a cause label and a handling plan;

[0009] Based on historical equipment failure events, a failure decision tree is constructed. The construction process of the failure decision tree specifically includes:

[0010] The time series data is divided into windows, and features are extracted from the time series data in each window to form a feature vector;

[0011] Cluster all feature vectors and map each cluster result to a state symbol, thereby transforming each historical equipment failure event into a symbolic sequence representing its state change process;

[0012] The symbolic sequences with the same causal labels are grouped into the same group to obtain the grouping results;

[0013] Using the normal state of all historical equipment failure events as the root node, the nodes of each layer in each decision branch are generated iteratively. Each decision branch contains at least one layer arranged in chronological order, and each decision branch corresponds to a grouping result of a cause label. Each layer of nodes includes a representation branch node for characterizing the frequency of symbols within a group, and a distinction branch node for characterizing the ability of symbols to distinguish between different groups. Each node is set with a risk value that characterizes the degree of risk of its equipment failure.

[0014] The system acquires real-time equipment fault data during equipment operation, converts the real-time equipment fault data into real-time symbol sequences, matches the paths corresponding to the real-time symbol sequences from the fault decision tree to obtain the target path, calculates the equipment fault risk probability based on the risk values ​​associated with the branch nodes on the target path, and outputs corresponding early warning information and fault handling measures based on the target path when the equipment fault risk probability exceeds the preset risk threshold.

[0015] Optionally, the time series data is divided into windows, and features are extracted from the time series data in each window to form a feature vector, specifically including:

[0016] For each historical device failure event, use a window to slide across all types of time-series data;

[0017] Feature extraction is performed on each type of time series data within the window to obtain the data features of each type of time series data;

[0018] By concatenating different types of data features, the feature vector of the time series data within the window is obtained;

[0019] The window length is a first preset length value, the window sliding length is a second preset length value, and the types of time-series data include at least two of temperature, pressure, vibration, and current.

[0020] Feature extraction is performed on each type of time series data within the window, specifically including:

[0021] Extract at least one of the mean, slope, and variance of the time series data within the window.

[0022] Optionally, the K-Means clustering algorithm is used to cluster all feature vectors.

[0023] Optionally, the normal state of all historical equipment failure events is used as the root node to iteratively generate nodes at each level in each decision branch, specifically including:

[0024] For any layer in any decision branch, obtain the symbols of all symbolized sequences in the grouping results corresponding to that decision branch at that layer, forming a candidate symbol set, wherein the order of the symbols in the corresponding symbolized sequences is the same as the layer number;

[0025] The frequency of each symbol in the candidate symbol set is counted, and all symbols are sorted in ascending order of frequency to obtain the sorted sequence;

[0026] The sorted sequence is segmented using the Otsu thresholding method to obtain a segmentation threshold. Symbols in the sorted sequence with a frequency greater than the segmentation threshold are taken as the representation branch nodes of that layer in the decision branch, and the representative value of each representation branch node is calculated.

[0027] Obtain all symbols in the candidate symbol set. For any symbol, obtain the number of times the symbol appears in different decision branches of the layer. The ratio of the number of occurrences to the total number of symbols in the corresponding decision branch is determined as the frequency of the symbol in different decision branches, thus obtaining the frequency sequence of the symbol.

[0028] The symbol is identified as the distinguishing branch node of the decision branch corresponding to the highest frequency in the frequency sequence, and the state distinguishing ability of each distinguishing branch node is calculated.

[0029] Optionally, the decision branches are constructed iteratively.

[0030] Decision branches are constructed iteratively, specifically including:

[0031] For the first layer of the decision branch, the symbol of this layer is the first symbol in each symbolization sequence;

[0032] For the Nth level of the decision branch, the specific methods for obtaining the symbols at this level include:

[0033] Determine the parent node of the layer, select the first N-1 symbols from the symbolized sequence of the group corresponding to the decision branch that have the same path to the parent node as the target symbolized sequence, and take the Nth symbol of the target symbolized sequence as the candidate symbol set of the Nth layer of the decision branch, where N is a positive integer.

[0034] Optionally, the representative value of each representative branch node is calculated, specifically including:

[0035] Calculate the mean frequency of symbols in the sorted sequence whose frequency is less than or equal to the segmentation threshold, and determine the representativeness of any representative branch node as the difference between the frequency of any representative branch node and the mean frequency.

[0036] Calculate the state discrimination capability of each distinguishing branch node, specifically including:

[0037] Obtain the frequency sequence of the distinguishing branch node, calculate the absolute value of the difference between any two frequencies in the frequency sequence, and average all the absolute values ​​to obtain the state distinguishing ability of the distinguishing branch node.

[0038] Optionally, the risk value can be calculated as follows:

[0039] Obtain historical equipment operation data;

[0040] From the symbol sequence converted from historical equipment operation data, select all M symbols that match the path from the root node to that node, and use them as the target historical sequence, where M is the number of layers of that node.

[0041] Calculate the proportion of the number of sequences in the target historical sequence whose cause label is equipment failure event, relative to the total number of all target historical sequences, and set this proportion as the risk value of that node.

[0042] Optionally, the target path is obtained by matching the path corresponding to the real-time symbol sequence from the fault decision tree, specifically including:

[0043] Set the root node of the fault decision tree as the initialization node and set an initial path;

[0044] The symbols in the real-time symbol sequence are matched sequentially with the child nodes of the initialization node in the fault decision tree;

[0045] If a match is successful, the matched child node is added to the initial path and updated to a new initialization node, and the matching of symbols in the real-time symbol sequence continues.

[0046] When the initialization node is a leaf node, or a symbol in the real-time symbol sequence cannot be successfully matched with any child node of the initialization node, or the temporary risk assessment value calculated based on the current initial path exceeds the preset warning threshold, the matching stops and the initial path is determined as the target path.

[0047] Optionally, the probability of equipment failure risk is calculated based on the risk values ​​associated with the branch nodes on the target path, specifically including:

[0048] For any branch node on the target path, if the branch node is both a representation branch node and a distinguishing branch node, the larger of the two risk values ​​associated with the node as a representation branch node and as a distinguishing branch node is used as the base value, and the state distinguishing ability of the node is increased by one as an improvement coefficient. The product of the base value and the improvement coefficient is calculated as the risk probability contribution value of the node.

[0049] If the branch node is a representative branch node, the larger of the representative value and the risk value of the representative branch node is determined as the risk probability contribution value of the node;

[0050] If the branch node is a distinguishing branch node, the larger of the distinguishing ability of the state and the risk value of the distinguishing branch node shall be determined as the risk probability contribution value of the node.

[0051] The maximum value of the risk probability contribution of all nodes on the target path is determined as the equipment failure risk probability.

[0052] This invention proposes an intelligent diagnostic processing system for equipment fault data, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the steps of an intelligent diagnostic processing method for equipment fault data.

[0053] The beneficial effects of the technical solution of the present invention are:

[0054] In this embodiment of the invention, data characteristics under normal equipment operation are used as the starting point (root node) for analysis. The growth direction of the tree structure strictly follows the time sequence, meaning that the path from the root node to each terminal leaf node simulates the complete process of the evolution of operating parameters over time caused by a typical equipment failure. Each path represents a sequence of data changes that gradually develops from a normal state to an equipment failure state due to a specific cause.

[0055] By matching real-time collected equipment operation data with this decision tree, the probability of equipment failure under current operating conditions can be dynamically assessed. As real-time data continues to be input, this probability value will be dynamically updated, thus enabling the tracking of risk evolution. More importantly, while assessing risk, fault handling measures that best match the current risk state and have been proven effective in similar historical situations can be generated simultaneously, thereby achieving rapid response from early warning to precise decision-making. Attached Figure Description

[0056] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0057] Figure 1 A flowchart illustrating an intelligent diagnostic processing method for equipment fault data provided in one embodiment of the present invention;

[0058] Figure 2 This is a structural diagram of an intelligent diagnostic and processing system for equipment fault data provided in one embodiment of the present invention. Detailed Implementation

[0059] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of an intelligent diagnostic processing method for equipment fault data proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0060] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0061] The following description, in conjunction with the accompanying drawings, details the specific scheme of the intelligent diagnostic and processing method for equipment fault data provided by the present invention.

[0062] This invention provides an intelligent diagnostic processing method and system for equipment fault data. Please refer to [link / reference]. Figure 1 The diagram illustrates a flowchart of an intelligent diagnostic processing method for equipment fault data according to an embodiment of the present invention. The method includes the following steps:

[0063] S101. Obtain historical equipment failure events, wherein the historical equipment failure events include at least two types of time-series data, and each historical equipment failure event corresponds to a cause label and a handling plan.

[0064] For example, in industrial scenarios, the acquisition of equipment-related data is actively performed by a data acquisition system according to a preset frequency or event-triggered mechanism. The system issues commands to various sensors, programmable logic controllers (PLCs), or distributed control systems (DCS) to acquire real-time operating parameters and status information of the equipment, such as temperature, pressure, vibration, and current. For automated equipment from different manufacturers and using different protocols, the system actively performs protocol parsing and data interfacing by configuring corresponding drive and communication modules, periodically polling or reading process data from the controller's memory in real time. The acquired raw data undergoes preprocessing steps such as noise reduction, cleaning, and time alignment to form a standardized data foundation that can be used for subsequent analysis.

[0065] A historical equipment failure event includes at least two types of time-series data such as equipment temperature, pressure, vibration, and current. Each historical equipment failure event corresponds to a cause of equipment failure and a corresponding solution for resolving the equipment failure.

[0066] S102. Based on historical equipment failure events, construct a failure decision tree. The construction process of the failure decision tree specifically includes:

[0067] The time series data is divided into windows, and features are extracted from the time series data in each window to form a feature vector;

[0068] Cluster all feature vectors and map each cluster result to a state symbol, thereby transforming each historical equipment failure event into a symbolic sequence representing its state change process;

[0069] The symbolic sequences with the same causal labels are grouped into the same group to obtain the grouping results;

[0070] Using the normal state of all historical equipment failure events as the root node, the nodes of each layer in each decision branch are generated iteratively. Each decision branch contains at least one layer arranged in chronological order, and each decision branch corresponds to a grouping result of a cause label. Each layer of nodes includes a representation branch node for characterizing the frequency of symbols within a group, and a differentiation branch node for characterizing the ability of symbols to distinguish between different groups. Each node is set with a risk value that characterizes the degree of risk of its equipment failure.

[0071] In this embodiment, the time series data is divided into windows, and features are extracted from the time series data in each window to form a feature vector, specifically including:

[0072] For each historical device failure event, use a window to slide across all types of time-series data;

[0073] Feature extraction is performed on each type of time series data within the window to obtain the data features of each type of time series data;

[0074] By concatenating different types of data features, the feature vector of the time series data within the window is obtained;

[0075] The window length is a first preset length value, the window sliding length is a second preset length value, and the types of time-series data include at least two of temperature, pressure, vibration, and current.

[0076] Feature extraction is performed on each type of time series data within the window, specifically including:

[0077] Extract at least one of the mean, slope, and variance of the time series data within the window;

[0078] The K-Means clustering algorithm is used to cluster all feature vectors.

[0079] Using the normal state of all historical equipment failure events as the root node, the nodes of each layer in each decision branch are iteratively generated, specifically including:

[0080] For any layer in any decision branch, obtain the symbols of all symbolized sequences in the grouping results corresponding to that decision branch at that layer, forming a candidate symbol set, wherein the order of the symbols in the corresponding symbolized sequences is the same as the layer number;

[0081] The frequency of each symbol in the candidate symbol set is counted, and all symbols are sorted in ascending order of frequency to obtain the sorted sequence;

[0082] The sorted sequence is segmented using the Otsu thresholding method to obtain a segmentation threshold. Symbols in the sorted sequence with a frequency greater than the segmentation threshold are taken as the representation branch nodes of that layer in the decision branch, and the representative value of each representation branch node is calculated.

[0083] Obtain all symbols in the candidate symbol set. For any symbol, obtain the number of times the symbol appears in different decision branches of the layer. The ratio of the number of occurrences to the total number of symbols in the corresponding decision branch is determined as the frequency of the symbol in different decision branches, thus obtaining the frequency sequence of the symbol.

[0084] The symbol is identified as the distinguishing branch node of the decision branch corresponding to the highest frequency in the frequency sequence, and the state distinguishing ability of each distinguishing branch node is calculated.

[0085] The decision branches are constructed iteratively.

[0086] Decision branches are constructed iteratively, specifically including:

[0087] For the first layer of the decision branch, the symbol of this layer is the first symbol in each symbolization sequence;

[0088] For the Nth level of the decision branch, the specific methods for obtaining the symbols at this level include:

[0089] Determine the parent node of the layer, select the first N-1 symbols from the symbolized sequence of the group corresponding to the decision branch that have the same path to the parent node as the target symbolized sequence, and take the Nth symbol of the target symbolized sequence as the candidate symbol set of the Nth layer of the decision branch, where N is a positive integer.

[0090] Calculate the representative value for each representative branch node, specifically including:

[0091] Calculate the mean frequency of symbols in the sorted sequence whose frequency is less than or equal to the segmentation threshold, and determine the representativeness of any representative branch node as the difference between the frequency of any representative branch node and the mean frequency.

[0092] Calculate the state discrimination capability of each distinguishing branch node, specifically including:

[0093] Obtain the frequency sequence of the distinguishing branch node, calculate the absolute value of the difference between any two frequencies in the frequency sequence, and average all the absolute values ​​to obtain the state distinguishing ability of the distinguishing branch node.

[0094] The risk value is calculated as follows:

[0095] Obtain historical equipment operation data;

[0096] From the symbol sequence converted from historical equipment operation data, select all M symbols that match the path from the root node to that node, and use them as the target historical sequence, where M is the number of layers of that node.

[0097] Calculate the proportion of the number of sequences in the target historical sequence whose cause label is equipment failure event, relative to the total number of all target historical sequences, and set this proportion as the risk value of that node.

[0098] Constructing a fault decision tree requires the systematic collection and processing of historical equipment fault event data. The specific steps are as follows:

[0099] First, define a "historical equipment failure event." This refers to a real equipment failure case that occurred in history and was ultimately resolved through effective fault handling measures. For each case, extract the complete data segment from when the equipment's operating parameters were completely normal until the equipment failure was identified and resolved. This data segment constitutes an independent sample of "historical equipment failure events."

[0100] Each such historical sample contains multiple high-frequency time-series data, mainly including: temperature time-series data, pressure time-series data, vibration time-series data, and current time-series data. Simultaneously, two key labels are recorded for each sample: one is the "causation label R," characterizing the root cause of the equipment failure, and the other is the "treatment plan S" successfully applied in this case.

[0101] Subsequently, features are extracted from various types of time-series data in each historical event. A sliding window method is used to segment each type of time-series data, resulting in a series of time-continuous windows. For the time-series data within each window, a set of statistical features is calculated, typically including the mean, slope, and variance of the data within that window.

[0102] Next, feature fusion across data dimensions is performed. The mean, slope, variance, and other features calculated from different types of time-series data (e.g., temperature and pressure) within the same window are concatenated in a predetermined order to form a high-dimensional feature vector. This vector comprehensively represents the overall operating status of the equipment system within this specific time window.

[0103] For example, consider a sliding window containing the temperature data sequence: 50, 51, 49, 52, 48. The calculated mean is 50, the variance (or standard deviation) is approximately 1.58, and the slope is -0.2, indicating a slight downward trend in temperature.

[0104] The pressure data sequence within the same window is: 20, 21, 22, 23, 24. The calculated mean is 22, the variance is approximately 1.58, and the slope is 1.0, indicating a stable upward trend in pressure.

[0105] By concatenating the feature values ​​of the above temperature and pressure data in sequence, the feature vector of this window is obtained as: 50, 1.58, −0.2, 22, 1.58, 1.0. A complete historical equipment failure event contains multiple sliding windows arranged in chronological order, thus generating a series of feature vectors arranged in chronological order.

[0106] A sliding window is a data segmentation technique with a fixed time length. It continuously slides across time-series data at a preset "window length" (i.e., a first preset length value), with the distance slid at each step called the "sliding length" (i.e., a second preset length value). Each window captures a segment of continuous data points for calculating local features. In this way, long, continuous time-series signals can be transformed into a series of overlapping or continuous, fixed-dimensional local data fragments, thereby capturing local patterns and evolutionary trends of the data over time. The length of the first preset length value can be set according to actual needs, such as five minutes or ten minutes.

[0107] Through sliding window and feature extraction, each historical equipment failure event data is transformed into a series of feature vectors arranged in chronological order.

[0108] Next, a comprehensive analysis is performed on all feature vectors generated from all historical events. The K-Means clustering algorithm is used to perform unsupervised clustering of these high-dimensional feature vectors. After the algorithm completes, all feature vectors are divided into several clusters (or categories). Each cluster is assigned a unique identifier, such as using uppercase letters A, B, C or numbers 1, 2, 3.

[0109] Each symbol corresponds to a specific equipment system state mode. For example, symbol A may represent "normal equipment operation state". It belongs to the feature vector of this cluster. Its numerical mode is usually characterized by: various operating parameters (such as pressure, temperature, vibration, current, etc.) being stable within the normal range without significant fluctuations, and all parameters operating smoothly within the safety threshold.

[0110] The symbol B may represent "minor equipment malfunction symptoms". The characteristic pattern of this cluster is: slight, persistent abnormal changes in certain parameters (such as a small drop in pressure or a slow rise in temperature), but which have not yet triggered a serious alarm. This is an early warning state that requires close monitoring.

[0111] Through the clustering and symbol mapping described above, each sliding window's corresponding feature vector is assigned a specific state symbol. By arranging the symbols sequentially according to the chronological order of these windows in the original historical data, a complete historical equipment failure event can be transformed into a "symbolic sequence" composed of state symbols. This sequence visually depicts the complete process of the equipment system's state evolving from normal to abnormal during a single equipment failure event.

[0112] At this point, the historical data preprocessing and state abstraction are complete.

[0113] Using a sample library of "symbolic sequences" labeled with causal tags (R) and treatment options (S), a "fault decision tree" is constructed for real-time diagnosis and decision-making. This tree will serve as a reference model for real-time monitoring data matching and inference.

[0114] After obtaining the symbolic sequences of all historical equipment failure events, the fault decision tree for real-time matching is constructed. The construction process employs an iterative generation method that combines intra-group representativeness with inter-group discriminability.

[0115] First, initialization is performed. A root node is created, representing the initial normal state of all historical device failure events. In the initial phase of construction, the symbolic sequence of all historical events is associated with this root node.

[0116] Subsequently, historical events are pre-grouped based on their key attributes. Since equipment failure events with the same root cause (i.e., the same cause label R) often exhibit high similarity in their development paths and effective handling solutions, all historical events are first divided into different groups according to their cause label R before constructing the decision tree. This grouping information will be fully considered in the subsequent node generation process, aiming to construct targeted decision branches for each cause group.

[0117] The goal of building a decision tree is to balance the following two key factors in the generation of nodes at each level to optimize the overall performance of the tree:

[0118] Representativeness within a group: The goal is for each decision branch generated (i.e., a complete path from the root node to a leaf node) to reflect and represent, as much as possible, the typical development pattern of equipment failure events within the specific cause group it corresponds to.

[0119] Inter-layer distinguishability: The goal is for different nodes existing side-by-side within the same layer to have sufficiently large differences. This allows for clear and rapid differentiation of different potential development paths at an early stage when the symbol sequence of real-time data is matched layer by layer with the decision tree, thereby accelerating the matching process and improving the accuracy of path prediction.

[0120] The generation of the first-level nodes in a decision tree begins with the analysis of the initial states of historical events. The specific process is as follows:

[0121] Extract the first symbol (i.e., the first element of each sequence) from the symbolized sequences corresponding to all historical equipment failure events. Since the first symbol in each symbolized sequence represents the equipment system state corresponding to the event at the start time (or the first time window), arranging these first symbols according to the order of their historical events forms a special sequence, which can be called the "first-level candidate sequence". Importantly, when forming this sequence, each symbol still retains the original cause label R grouping information of its historical event.

[0122] For example, suppose there are five historical equipment failure events, and their symbolic sequences are as follows: Sequence 1: [x11, x12, x13, x14, x15...], Sequence 2: [x21, x22, x23, x24, x25...], Sequence 3: [x31, x32, x33, x34, x35...], Sequence 4: [x41, x42, x43, x44, x45...], Sequence 5: [x51, x52, x53, x54, x55...].

[0123] Sequence 1 and Sequence 2 belong to the same cause group (e.g., "bearing wear failure"), while Sequences 3, 4, and 5 belong to another cause group (e.g., "motor overheating failure").

[0124] Therefore, the extracted "first-level candidate sequence" is: [x11, x21, x31, x41, x51]. In this sequence, symbols x11 and x21 belong to the first causal group, and symbols x31, x41, and x51 belong to the second causal group. Based on this "first-level candidate sequence" with grouping information, and according to the aforementioned principles of intra-group representativeness and inter-group distinguishability, the representative branch nodes and distinguishing branch nodes of each causal group are selected and determined, thus forming the first-level structure of the decision tree.

[0125] At each level of the decision tree, a suitable symbol needs to be selected as a splitting node for each causal group. The selection criteria are based on two core indicators: first, the frequency of the symbol's occurrence within its group; the higher the frequency, the more representative the symbol is of the typical initial or early state of the group, i.e., the stronger its intra-group representation; second, the symbol's ability to distinguish between different groups; the stronger the distinguishing ability, the more effectively the symbol can differentiate the current group from other groups, thus enabling early and rapid identification in real-time matching.

[0126] To ensure that the decision tree possesses both good internal representation and external discriminative power, and provides a clear path logic for real-time matching, this embodiment designs two branches with different functional orientations for each cause group: a representation branch (analogous to the left branch) and a discriminative branch (analogous to the right branch).

[0127] Representational branches: Their main function is to capture and represent the most typical and common state evolution patterns of the causal group. When selecting nodes for this branch, the frequency of symbols appearing within the group is the primary consideration, with higher-frequency symbols being prioritized. If an early symbol of a real-time event matches a representational branch of a group, it suggests that the event is highly likely to be developing along the typical path of that group.

[0128] Differentiating branches: Their main function is to enhance the distinction between this group and other groups in the early stages. When selecting nodes for this branch, the primary consideration is the inter-group distinguishability of the symbols, prioritizing symbols that maximize differentiation from other groups. This helps in handling events with atypical initial states but highly identifiable characteristics, enabling them to be correctly classified in the early stages.

[0129] It's important to clarify that "left branch" (representation branch) and "right branch" (discrimination branch) are conceptual divisions of node function, not a rule that each group can only have two splitting nodes. Depending on the data distribution, a group may have multiple high-frequency symbols, resulting in multiple "representation branches"; simultaneously, it may have one or more highly discriminative symbols, resulting in one or more "discrimination branches." These two types of branches together constitute multiple child nodes originating from the parent node, but they are assigned different logical roles to collaboratively achieve accurate state recognition and path prediction.

[0130] The process of determining the characterization and differentiation of branches is as follows:

[0131] Representing branch nodes (functional analogy to the left branch):

[0132] Selection criteria: Among all candidate symbols in the current layer, the symbol with the highest frequency in the statistical target causal group is selected as the representation branch node. If multiple symbols have the same frequency and are all the highest, then all of them can be established as representation branch nodes for that group.

[0133] Function: This branch aims to quickly capture and characterize the most typical and common starting point of state evolution for this cause group. When the first symbol in the real-time monitoring data transformation sequence successfully matches the characterization branch node of a cause group, it indicates that the current operating condition is highly likely to be developing along the typical equipment failure mode represented by that group.

[0134] Distinguish branch nodes (functional analogy to the right branch):

[0135] Selection Criteria: The core criterion for selecting this node is the strong inter-group discriminative power of the symbol. The distribution of each candidate symbol across all causal groups is calculated, and its "state discriminative power" is evaluated by quantifying the differences in its distribution across different groups (e.g., calculating the mean difference in the symbol's frequency across groups). The symbol with the highest state discriminative power is selected as the discriminative branch node for that group. Similarly, if multiple symbols have the same and highest discriminative power, all of them can be selected.

[0136] Function: This branch is designed to handle events that have atypical initial states but distinct characteristics. By selecting symbols whose occurrence in this group differs significantly from other groups, it can clearly distinguish this group from other groups early in the decision tree process, thus ensuring that even if real-time data starts in an atypical state, it can be correctly guided to the corresponding analysis path.

[0137] The generation of first-level nodes follows these specific steps:

[0138] Step 1: Determine the representative branch nodes for each causal group:

[0139] First, for the target causal group, extract all symbols in the "first-level candidate sequence" to form the "intra-group candidate symbol sequence". Calculate the frequency of each symbol in this sequence.

[0140] Then, these symbols are sorted in ascending order of their frequency values ​​to obtain a "frequency ascending sequence".

[0141] For this ascending frequency sequence, the Otsu thresholding algorithm is applied to automatically calculate an optimal segmentation threshold K. The goal of this algorithm is to find a threshold that minimizes the intra-class variance and maximizes the inter-class variance between the two classes (high-frequency class and low-frequency class) divided by this threshold.

[0142] Symbols in the ascending frequency sequence whose frequency values ​​are greater than the segmentation threshold K are identified as the representative branch nodes of the causal group.

[0143] Each representation branch node is recorded in the form of (element symbol, representativeness). The "representativeness" is calculated as follows: first, the average frequency of all symbols less than or equal to the segmentation threshold K in the ascending frequency sequence is calculated; then, this average value is subtracted from the frequency value of the representation branch node itself, and the difference is its "representativeness". The higher the representativeness value, the more prominent and typical the node symbol is compared to the general low-frequency symbols in the group.

[0144] Step 2: Calculate the state discrimination capability of each symbol and assign it as a discriminative branch node:

[0145] To assess the inter-group discrimination potential of all symbols, a global calculation is performed: for each symbol in the "first-level candidate sequence", its frequency of occurrence in each causal group is counted, thus obtaining a "frequency sequence" describing the cross-group distribution of the symbol.

[0146] To quantify its discriminative power, the "state discriminative power" of the frequency sequence is calculated. Specifically, the absolute value of the difference between any two elements in the frequency sequence is calculated, and then the average of all these differences is taken. The larger this average value, the more uneven the distribution of the symbol across different groups, and the stronger its state discriminative power. Through this calculation, the state discriminative power value for each symbol is obtained.

[0147] Key logic (node ​​allocation): After obtaining the state discrimination capability value of each symbol, the following rules are followed for allocation: Each symbol is assigned to the cause group (or those groups) with the highest state discrimination capability value, as the distinguishing branch node of that group.

[0148] For example, suppose symbol A has state discrimination power values ​​calculated in the six causal groups as [0.1, 0.2, 0.2, 0.3, 0.3, 0.3]. Then, symbol A will be simultaneously established as a discrimination branch node in the fourth, fifth, and sixth causal groups. Each discrimination branch node is recorded in the form of (element symbol, state discrimination power), where "state discrimination power" is the highest discrimination power value achieved by that symbol in the corresponding group.

[0149] Step 3: Iterate through and complete the first layer construction:

[0150] For each causal group, steps one and two above are repeated. Through this process, each group obtains a representation branch node based on its own data characteristics (recorded as (element symbol, representativeness)) and a distinguishing branch node based on the global symbol distinguishing ability (recorded as (element symbol, state distinguishing ability)).

[0151] Through the above two steps, two types of nodes are generated for each causal group in the first layer of the decision tree: "representation branch nodes" selected based on intra-group frequency and "discriminative branch nodes" assigned based on global discriminative ability. These two types of nodes work together to optimize the decision tree's ability to identify and route real-time data early from different perspectives.

[0152] The generation of nodes in the second and subsequent levels of the decision tree uses the same core logic as the first level, employing iterative calculations. The specific process is as follows:

[0153] Generation of second-level nodes:

[0154] Take any node in the first layer (denoted as node A) as the "parent node" or "initial node" of the current layer. First, select all historical events that belong to the causal group of node A and whose first symbol in the symbolization sequence matches the symbol of node A. Define these events as the "current event set".

[0155] For the current set of events, extract the second symbol from the symbolic sequence of all events to form the "second-level candidate sequence" under this node.

[0156] Then, this second-level candidate sequence is processed entirely in the same way as the first-level nodes:

[0157] Generate left branch (representation branch node): Based on the method in [Step 1] (i.e., statistical group frequency, sorting, Otsu threshold segmentation, and representativeness calculation), generate a representation branch node for this node in the current layer and record it as (element symbol, representativeness).

[0158] Generate right branch (distinguishing branch node): Based on the method in [Step 2] (i.e., calculate the state discrimination ability of each symbol and assign the symbol to the group with the highest discrimination ability), generate a distinguishing branch node for the node in the current layer and record it as (element symbol, state discrimination ability).

[0159] Higher-level iterative construction:

[0160] The nodes in the third, fourth, and all subsequent layers repeat the generation process described in the second layer. Each layer takes a node from the previous layer as its parent node, selects a subset of historical events whose symbol sequences match this path, extracts the symbols at the corresponding positions to form new candidate sequences, and then applies the same two-step method (frequency-based representation branch generation and discrimination branch allocation based on discrimination ability) to split and generate new child nodes.

[0161] Through this layer-by-layer iterative, path-dependent splitting process, a complete fault decision tree is eventually constructed. Each leaf node of the decision tree corresponds to an effective handling plan S learned from historical data.

[0162] In a decision tree, each complete path from the root node (initial normal state) to a certain leaf node (specific handling plan) represents a typical equipment failure development pattern.

[0163] To quantify risk, the decision tree assigns a risk value to each node, representing the probability of equipment failure when real-time data matches and reaches that node. One approach is to set the risk value of leaf nodes to 1 (indicating that action is required) and then backtrack along the path, manually assigning an increasing risk value to each intermediate node along the path.

[0164] A more objective approach is based on historical statistics: For a given node, collect all historical data on events that have reached that node's state (including those that ultimately resulted in equipment failure and those that did not), and then calculate the proportion of events that ultimately resulted in equipment failure to the total number of events. This proportion is the node's risk value, which objectively reflects the probability of an accident occurring when in that state from a data perspective.

[0165] S103. Obtain real-time equipment fault data during equipment operation, convert the real-time equipment fault data into a real-time symbol sequence, match the path corresponding to the real-time symbol sequence from the fault decision tree to obtain the target path, calculate the equipment fault risk probability based on the risk value associated with the branch nodes on the target path, and output the corresponding early warning information and fault handling measures based on the target path when the equipment fault risk probability exceeds the preset risk threshold.

[0166] In this embodiment, the target path is obtained by matching the path corresponding to the real-time symbol sequence from the fault decision tree, specifically including:

[0167] Set the root node of the fault decision tree as the initialization node and set an initial path;

[0168] The symbols in the real-time symbol sequence are matched sequentially with the child nodes of the initialization node in the fault decision tree;

[0169] If a match is successful, the matched child node is added to the initial path and updated to a new initialization node, and the matching of symbols in the real-time symbol sequence continues.

[0170] When the initialization node is a leaf node, or a symbol in the real-time symbol sequence cannot be successfully matched with any child node of the initialization node, or the temporary risk assessment value calculated based on the current initial path exceeds the preset warning threshold, the matching stops and the initial path is determined as the target path.

[0171] The probability of equipment failure risk is calculated based on the risk values ​​associated with the branch nodes on the target path, specifically including:

[0172] For any branch node on the target path, if the branch node is both a representation branch node and a distinguishing branch node, the larger of the two risk values ​​associated with the node as a representation branch node and as a distinguishing branch node is used as the base value, and the state distinguishing ability of the node is increased by one as an improvement coefficient. The product of the base value and the improvement coefficient is calculated as the risk probability contribution value of the node.

[0173] If the branch node is a representative branch node, the larger of the representative value and the risk value of the representative branch node is determined as the risk probability contribution value of the node;

[0174] If the branch node is a distinguishing branch node, the larger of the distinguishing ability of the state and the risk value of the distinguishing branch node shall be determined as the risk probability contribution value of the node.

[0175] The maximum value of the risk probability contribution of all nodes on the target path is determined as the equipment failure risk probability.

[0176] For example, the preset risk thresholds include a first risk threshold and a second risk threshold. When the probability of equipment failure exceeds the preset risk threshold, corresponding early warning information and fault handling measures are output according to the target path, specifically including:

[0177] When the probability of equipment failure exceeds the first risk threshold but does not exceed the second risk threshold, a medium-risk warning message and general intervention suggestions are output.

[0178] When the probability of equipment failure exceeds the second risk threshold, a high-risk warning message and specific failure handling measures associated with the leaf nodes of the target path are output.

[0179] After the fault decision tree is constructed, it can be used for online analysis and decision-making based on real-time equipment operation monitoring data. This process mainly includes three stages: real-time data symbolization processing, decision tree path matching, and dynamic risk assessment.

[0180] Real-time data symbolization processing:

[0181] For continuously collected real-time multivariate time-series data (such as temperature, pressure, vibration, current, etc.), the same sliding window parameters (window length, sliding length) and feature extraction methods (calculating mean, slope, variance) as historical data preprocessing are used to generate a series of real-time feature vectors.

[0182] Subsequently, using a pre-trained clustering model (such as K-Means), each real-time feature vector is mapped to its corresponding state symbol. Finally, the real-time data is transformed into a sequence of real-time symbols arranged in chronological order, such as [S1, S2, S3, ...].

[0183] Tree path matching and dynamic risk assessment:

[0184] The system starts from the root node of the decision tree and performs layer-by-layer matching and risk assessment of the real-time symbol sequence with the decision tree, which is a dynamic and progressive process. During the process, judgments are made based on two preset risk thresholds: the lower one is called the first risk threshold (e.g., 0.4), which represents a medium risk level that requires vigilance and initial intervention; the higher one is called the second risk threshold (e.g., 0.7), which represents a high risk that requires immediate initiation of substantive emergency preparedness actions.

[0185] Optionally, the first risk threshold and the second risk threshold can be set and adjusted according to the actual situation. In this embodiment, they can be set to 0.4 and 0.7 respectively, without any specific restrictions.

[0186] Initialization: Set the current node as the root node, and initialize a current path and a current overall risk value.

[0187] Layer-by-layer matching: The symbols in the real-time symbol sequence are matched sequentially with all child nodes of the current node.

[0188] Risk assessment logic:

[0189] If the matched node is only a branch node, then its associated risk value is taken as the contribution of this match.

[0190] If the matched node is only a distinguishing branch node, then its associated risk value is taken as the contribution.

[0191] Key scenario: If the matched node is both a representational branch node and a distinguishing branch node, the higher risk value is used as the base, and its state distinguishing ability is used as an enhancement coefficient. The risk contribution value of the current node is: higher risk value × (1 + state distinguishing ability). This amplifies the weight of high-distinguishability nodes in risk assessment.

[0192] Furthermore, an incremental accumulation or maximum value strategy is adopted to integrate the risk contribution value of this match into the current comprehensive risk value.

[0193] Tiered early warning and action:

[0194] When the current overall risk value exceeds the first risk threshold (e.g., 0.4) for the first time, the system triggers a medium-risk warning. At this point, the matching depth may still be shallow and the path may not be fully clear. Therefore, it is recommended to prepare key intervention measures necessary for all possible solutions under this path (such as preparing spare parts, adjusting operating parameters, and arranging shutdown for maintenance) in advance, and to closely monitor subsequent data.

[0195] Matching continues. When the current overall risk value further increases and exceeds the second risk threshold (e.g., 0.7), the system triggers a high-risk warning. At this point, emergency preparedness is immediately initiated, core handling measures under this path are executed (such as emergency shutdown and replacement of critical components), and the final handling plan is determined based on the matched path.

[0196] Matching terminates: Matching stops when one of the following occurs: the real-time symbol sequence matches a leaf node, the next symbol in the real-time symbol sequence cannot match any child node of the current node, or the real-time symbol sequence has finished matching.

[0197] Output results:

[0198] After the matching terminates, the system outputs the complete analysis results:

[0199] Predicted causes of equipment failure: Based on the cause group to which the final matched path belongs, the most likely cause of equipment failure is given.

[0200] Development path probability assessment: Provides the state probability of key nodes on the currently matched path, comprehensively reflecting the credibility of the current diagnostic results.

[0201] Output: If a leaf node is matched, the historically successful handling scheme S associated with that leaf node is directly recommended; if the process terminates early, the most likely handling scheme pointed to by the current path is recommended, or the optimal suggestion is given based on the comprehensive path information.

[0202] In summary, this invention, through the transformation of high-dimensional equipment runtime sequence data into a sequence of state symbols and the construction of a fault decision tree that combines intra-group representativeness and inter-group discriminability, achieves dynamic simulation and early identification of equipment fault development paths. This method can issue early warnings when equipment fault symptoms first appear, dynamically assess risk probabilities based on real-time data input, and automatically recommend historically validated and accurate fault handling solutions when the risk exceeds a threshold. This effectively overcomes the limitations of traditional threshold alarms, which are characterized by lag and reliance on human experience, improving the timeliness, objectivity, and success rate of equipment fault response, and providing intelligent decision support for safe equipment operation.

[0203] This invention also proposes an intelligent diagnostic and processing system for equipment fault data; please refer to [link / reference]. Figure 2 The diagram shows a structural diagram of an intelligent diagnostic processing system for equipment fault data provided in an embodiment of the present invention. The system includes: a data acquisition module 101, a decision tree construction module 102, and a fault early warning module 103.

[0204] The data acquisition module 101 is used to acquire historical equipment failure events, wherein the historical equipment failure events include at least two types of time-series data, and each historical equipment failure event corresponds to a cause label and a handling plan;

[0205] The decision tree construction module 102 is used to construct a fault decision tree based on historical equipment failure events. The specific process of constructing the fault decision tree includes:

[0206] The time series data is divided into windows, and features are extracted from the time series data in each window to form a feature vector;

[0207] Cluster all feature vectors and map each cluster result to a state symbol, thereby transforming each historical equipment failure event into a symbolic sequence representing its state change process;

[0208] The symbolic sequences with the same causal labels are grouped into the same group to obtain the grouping results;

[0209] Using the normal state of all historical equipment failure events as the root node, the nodes of each layer in each decision branch are generated iteratively. Each decision branch contains at least one layer arranged in chronological order, and each decision branch corresponds to a grouping result of a cause label. Each layer of nodes includes a representation branch node for characterizing the frequency of symbols within a group, and a distinction branch node for characterizing the ability of symbols to distinguish between different groups. Each node is set with a risk value that characterizes the degree of risk of its equipment failure.

[0210] The fault early warning module 103 is used to acquire real-time equipment fault data during equipment operation, convert the real-time equipment fault data into real-time symbol sequences, match the paths corresponding to the real-time symbol sequences from the fault decision tree to obtain the target path, calculate the equipment fault risk probability based on the risk values ​​associated with the branch nodes on the target path, and output corresponding early warning information and fault handling measures based on the target path when the equipment fault risk probability exceeds the preset risk threshold.

[0211] It should be noted that the system provided in the above embodiments is only an example of the division of the above functional modules. In practical applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the computer device can be divided into different functional modules to complete all or part of the functions described above. In addition, the intelligent diagnostic processing system for equipment fault data and the intelligent diagnostic processing method for equipment fault data provided in the above embodiments belong to the same concept, and their specific implementation process can be found in the method embodiments, which will not be repeated here.

[0212] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0213] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0214] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent diagnosis processing of equipment failure data, characterized in that, include: Acquire historical equipment failure events, wherein the historical equipment failure events include at least two types of time-series data, and each historical equipment failure event corresponds to a cause label and a handling plan; Based on historical equipment failure events, a failure decision tree is constructed. The construction process of the failure decision tree specifically includes: The time series data is divided into windows, and features are extracted from the time series data in each window to form a feature vector; Cluster all feature vectors and map each cluster result to a state symbol, thereby transforming each historical equipment failure event into a symbolic sequence representing its state change process; The symbolic sequences with the same causal labels are grouped into the same group to obtain the grouping results; Using the normal state of all historical equipment failure events as the root node, the nodes of each layer in each decision branch are generated iteratively. Each decision branch contains at least one layer arranged in chronological order, and each decision branch corresponds to a grouping result of a cause label. Each layer of nodes includes a representation branch node for characterizing the frequency of symbols within a group, and a distinction branch node for characterizing the ability of symbols to distinguish between different groups. Each node is set with a risk value that characterizes the degree of risk of its equipment failure. The system acquires real-time equipment fault data during equipment operation, converts the real-time equipment fault data into real-time symbol sequences, matches the paths corresponding to the real-time symbol sequences from the fault decision tree to obtain the target path, calculates the equipment fault risk probability based on the risk values ​​associated with the branch nodes on the target path, and outputs corresponding early warning information and fault handling measures based on the target path when the equipment fault risk probability exceeds the preset risk threshold.

2. The intelligent diagnosis processing method of equipment failure data according to claim 1, characterized in that, The process of dividing the time-series data into windows and extracting features from the time-series data in each window to form a feature vector specifically includes: For each historical device failure event, use a window to slide across all types of time-series data; Feature extraction is performed on each type of time series data within the window to obtain the data features of each type of time series data; By concatenating different types of data features, the feature vector of the time series data within the window is obtained; The window length is a first preset length value, the window sliding length is a second preset length value, and the types of time-series data include at least two of temperature, pressure, vibration, and current. Feature extraction is performed on each type of time series data within the window, specifically including: Extract at least one of the mean, slope, and variance of the time series data within the window.

3. The intelligent diagnostic processing method for equipment fault data according to claim 1, characterized in that, The algorithm used to cluster all feature vectors is the K-Means clustering algorithm.

4. The intelligent diagnostic processing method for equipment fault data according to claim 1, characterized in that, The process of using the normal state of all historical equipment failure events as the root node and iteratively generating nodes at each level in each decision branch specifically includes: For any layer in any decision branch, obtain the symbols of all symbolized sequences in the grouping results corresponding to that decision branch at that layer, forming a candidate symbol set, wherein the order of the symbols in the corresponding symbolized sequences is the same as the layer number; The frequency of each symbol in the candidate symbol set is counted, and all symbols are sorted in ascending order of frequency to obtain the sorted sequence; The sorted sequence is segmented using the Otsu thresholding method to obtain a segmentation threshold. Symbols in the sorted sequence with a frequency greater than the segmentation threshold are taken as the representation branch nodes of that layer in the decision branch, and the representative value of each representation branch node is calculated. Obtain all symbols in the candidate symbol set. For any symbol, obtain the number of times the symbol appears in different decision branches of the layer. The ratio of the number of occurrences to the total number of symbols in the corresponding decision branch is determined as the frequency of the symbol in different decision branches, thus obtaining the frequency sequence of the symbol. The symbol is identified as the distinguishing branch node of the decision branch corresponding to the highest frequency in the frequency sequence, and the state distinguishing ability of each distinguishing branch node is calculated.

5. The intelligent diagnostic processing method for equipment fault data according to claim 4, characterized in that, The decision branches are constructed iteratively. The iterative construction of decision branches specifically includes: For the first layer of the decision branch, the symbol of this layer is the first symbol in each symbolization sequence; For the Nth level of the decision branch, the specific methods for obtaining the symbols at this level include: Determine the parent node of the layer, select the first N-1 symbols from the symbolized sequence of the group corresponding to the decision branch that have the same path to the parent node as the target symbolized sequence, and take the Nth symbol of the target symbolized sequence as the candidate symbol set of the Nth layer of the decision branch, where N is a positive integer.

6. The intelligent diagnostic processing method for equipment fault data according to claim 4, characterized in that, The calculation of the representative value of each representative branch node specifically includes: Calculate the mean frequency of symbols in the sorted sequence whose frequency is less than or equal to the segmentation threshold, and determine the representativeness of any representative branch node as the difference between the frequency of any representative branch node and the mean frequency. Calculate the state discrimination capability of each distinguishing branch node, specifically including: Obtain the frequency sequence of the distinguishing branch node, calculate the absolute value of the difference between any two frequencies in the frequency sequence, and average all the absolute values ​​to obtain the state distinguishing ability of the distinguishing branch node.

7. The intelligent diagnostic processing method for equipment fault data according to claim 1, characterized in that, The risk value is calculated as follows: Obtain historical equipment operation data; From the symbol sequence converted from historical equipment operation data, select all M symbols that match the path from the root node to that node, and use them as the target historical sequence, where M is the layer number of that node. Calculate the proportion of the number of sequences in the target historical sequence whose cause label is equipment failure event, relative to the total number of all target historical sequences, and set this proportion as the risk value of that node.

8. The intelligent diagnostic processing method for equipment fault data according to claim 1, characterized in that, The step of matching the path corresponding to the real-time symbol sequence from the fault decision tree to obtain the target path specifically includes: Set the root node of the fault decision tree as the initialization node and set an initial path; The symbols in the real-time symbol sequence are matched sequentially with the child nodes of the initialization node in the fault decision tree; If a match is successful, the matched child node is added to the initial path, and the child node is updated to a new initialization node, and the matching of symbols in the real-time symbol sequence continues. When the initialization node is a leaf node, or a symbol in the real-time symbol sequence cannot be successfully matched with any child node of the initialization node, or the temporary risk assessment value calculated based on the current initial path exceeds the preset warning threshold, the matching stops and the initial path is determined as the target path.

9. The intelligent diagnostic processing method for equipment fault data according to claim 1, characterized in that, The calculation of equipment failure risk probability based on the risk values ​​associated with branch nodes on the target path specifically includes: For any branch node on the target path, if the branch node is both a representation branch node and a distinguishing branch node, the larger of the two risk values ​​associated with the node as a representation branch node and as a distinguishing branch node is used as the base value, and the state distinguishing ability of the node is increased by one as an improvement coefficient. The product of the base value and the improvement coefficient is calculated as the risk probability contribution value of the node. If the branch node is a representative branch node, the larger of the representative value and the risk value of the representative branch node is determined as the risk probability contribution value of the node; If the branch node is a distinguishing branch node, the larger of the distinguishing ability of the state and the risk value of the distinguishing branch node shall be determined as the risk probability contribution value of the node. The maximum value of the risk probability contribution of all nodes on the target path is determined as the equipment failure risk probability.

10. An intelligent diagnostic processing system for equipment fault data, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the computer program is executed by the processor, it implements the steps of the intelligent diagnostic processing method for equipment fault data as described in any one of claims 1-9.

Images