Method and system for identifying safety risk of equipment based on continuous interaction of mine agent

Abstract

The embodiment of the application provides a device safety risk identification method and system based on continuous interaction of mine agents, after the mine agent receives and analyzes device abnormal natural language description, calls analysis multi-source time sequence signal flow, outputs an initial fault hypothesis set and an initial fault confidence set, according to the confidence deviation of the initial fault confidence set relative to a preset confidence threshold, constructs a fault hypothesis space convergence strategy of the initial fault hypothesis set to call multi-source compensation signal flow and iteratively update fault root cause, until the updated fault confidence set meets the confidence threshold, and outputs a structured device fault diagnosis report. The problems that the existing mine device fault diagnosis exists cross-monitoring system data island and relies on artificial experience, which leads to deviation of fault diagnosis and delay of diagnosis result feedback response are solved. The mine device fault diagnosis achieves a second-level response, completely eliminates the artificial experience blind area and cross-system data island, and greatly reduces the operation and maintenance cost.

Description

Technical Field

[0001] This invention relates to the field of intelligent diagnostic technology for mining equipment, and in particular to a method and system for identifying equipment safety risks based on continuous interaction of mining intelligent agents. Background Technology

[0002] Existing technologies face significant bottlenecks in mine equipment fault diagnosis and inspection. This is mainly manifested in the fact that knowledge transfer relies heavily on the accumulation of human experience and cross-departmental communication, which prevents new employees from quickly acquiring key knowledge of equipment maintenance, resulting in a gap in capabilities.

[0003] Meanwhile, equipment operation data is scattered across multiple independent monitoring systems, requiring manual login to different platforms for repeated queries. On-site inspection efficiency is hampered by data retrieval delays and cumbersome operations. Furthermore, the lack of dynamic linkage between real-time data and historical cases in the fault diagnosis process leads to subjective blind spots in analysis and decision-making, resulting in biased diagnostic results. Moreover, the report writing process requires manually integrating fragmented information and adapting it to standardized formats, which is time-consuming and causes delays in the feedback of diagnostic results.

[0004] This fragmented operation of the entire fault diagnosis process, from knowledge acquisition and data retrieval to report output, relies heavily on manual intervention, exacerbating both diagnostic bias and delayed diagnostic response, thus creating a systemic efficiency bottleneck that restricts safe and efficient mine production.

[0005] It should be noted that the information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0006] To address the aforementioned deficiencies or improvement needs of existing technologies, this invention provides a method and system for identifying equipment safety risks based on continuous interaction of mining intelligent agents. This solves the problems in existing technologies where fault diagnosis of mining equipment suffers from data silos across monitoring systems and reliance on human experience, leading to diagnostic biases and delayed feedback. By using anomaly descriptions of mining equipment, the intelligent agent retrieves multi-source signal streams, fuses and analyzes them through a state machine, and outputs initial fault hypotheses and confidence levels. Based on the confidence deviation, compensation signals are collected and iteratively updated. Once a threshold is met, a structured diagnostic report containing a fault location matrix is ​​output, achieving a second-level response time for mining equipment fault diagnosis. This significantly shortens the diagnostic cycle and automatically generates executable structured reports, completely eliminating blind spots in human experience and cross-system data silos, thereby substantially reducing operation and maintenance costs. The specific technical solution is as follows:

[0007] According to a first aspect of the present invention, a method for identifying equipment safety risks based on continuous interaction of mining intelligent agents is provided, the method comprising:

[0008] Equipment maintenance personnel upload natural language descriptions of equipment anomalies via a mobile terminal text interface; the mining agent receives and parses these descriptions, identifying the equipment to be investigated and a preliminary list of anomaly types; based on the equipment to be investigated, the data retrieval engine is activated, synchronously retrieving multi-source time-series signal streams from multiple distributed equipment monitoring modules via a pre-defined interface protocol; after initializing the context state machine based on the preliminary anomaly type list, the mining agent uses the context state machine to fuse and analyze the multi-source time-series signal streams, outputting an initial fault hypothesis set and an initial fault confidence set; based on the confidence deviation of the initial fault confidence set relative to a pre-set confidence threshold, a fault hypothesis space convergence strategy is constructed for the initial fault hypothesis set; based on the fault hypothesis space convergence strategy, a directional data retrieval instruction is generated, driving the data retrieval engine to retrieve multi-source compensation signal streams; the multi-source compensation signal streams and multi-source time-series signal streams are input into the context state machine for progressive iteration of fault root causes until the updated fault confidence set meets the confidence threshold, and a structured equipment fault diagnosis report is output.

[0009] In one implementation, after the mining agent initializes the context state machine based on the preliminary anomaly type list, it uses the context state machine to perform fusion analysis on the multi-source time-series signal stream, outputs an initial fault hypothesis set and an initial fault confidence set, and then performs the following processing:

[0010] Based on the fault feature dimensions of the preliminary anomaly type list, directional feature extraction is performed on the multi-source time-series signal stream to construct a preliminary feature vector; a state machine instance is created based on the preliminary anomaly type list to obtain the context state machine; in the context state machine, the preliminary feature vector is used to traverse the structured fault case graph for similarity matching under multi-source information fusion to obtain a candidate fault pattern set; dynamic confidence decision is performed on the candidate fault pattern set to obtain the initial fault hypothesis set and the initial fault confidence set.

[0011] In one implementation, the following processing is also performed:

[0012] The multi-source time-series signal stream is subjected to continuity detection to locate the sequence of missing data nodes; the multi-source time-series signal stream is subjected to stability assessment to identify the sequence of abnormal fluctuation nodes; based on the sequence of missing data nodes and the sequence of abnormal fluctuation nodes, a reverse parameter retrieval instruction is constructed to obtain an incremental compensation data stream; the incremental compensation data stream is superimposed on the multi-source time-series signal stream to obtain a multi-source corrected signal stream; according to the fault feature dimension of the preliminary anomaly type list, the multi-source corrected signal stream is subjected to directional feature extraction to construct the preliminary feature vector.

[0013] In one implementation, dynamic confidence decision is performed on the candidate fault mode set to obtain the initial fault hypothesis set and the initial fault confidence set, and the following processing is also performed:

[0014] Based on the preliminary feature vector, the physical association strength weight coefficient of the fault feature dimension is determined, and the weighted similarity is calculated on the candidate fault pattern set to generate the initial fault hypothesis set. From the candidate fault pattern set, multiple historical case fits, multiple prior probability values ​​and multiple feature matching degrees of multiple fault hypotheses are extracted, and dynamic weighted fusion calculation is performed to output the initial fault confidence set.

[0015] In one implementation, based on the confidence deviation of the initial fault confidence set relative to a preset confidence threshold, a fault hypothesis space convergence strategy for the initial fault hypothesis set is constructed, and the following processing is further performed:

[0016] By sorting the confidence deviations of the initial fault confidence set relative to the confidence threshold in descending order, W low-confidence fault hypotheses are located in the initial fault hypothesis set. After matching the W sets of verification indicators of the W low-confidence fault hypotheses in the knowledge base, the missing evidence indicator set is obtained by indicator aggregation, which serves as the convergence strategy for the fault hypothesis space.

[0017] In one implementation, the multi-source compensation signal stream and the multi-source timing signal stream are input into the context state machine for progressive iteration of fault root causes until the updated fault confidence set satisfies the confidence threshold, and a structured equipment fault diagnosis report is output. The following processing is also performed:

[0018] Based on the missing evidence indicator set, a targeted data retrieval instruction is constructed to drive the data retrieval engine to retrieve multi-source compensation signal streams. According to the fault feature dimensions of the preliminary anomaly type list, targeted feature extraction is performed on the multi-source compensation signal stream and the multi-source time-series signal stream to construct an updated feature vector. In the context state machine, the updated feature vector is used to traverse the structured fault case graph to perform similarity matching under multi-source information fusion, resulting in an updated fault pattern set. Based on the updated feature vector, dynamic weight allocation is performed on the updated fault pattern set to generate an updated fault hypothesis set. Then, dynamic confidence decision-making is performed on the updated fault pattern set to output an updated fault confidence set. This process is repeated iteratively to iterate the root cause of the fault until the updated fault confidence set meets the confidence threshold, at which point a structured equipment fault diagnosis report is output.

[0019] In one implementation, the mining agent receives and parses the natural language description of the equipment anomaly, identifies the equipment objects to be investigated and a preliminary list of anomaly types, and then performs the following processing:

[0020] The mining agent invokes the built-in natural language processing module to perform part-of-speech tagging and entity boundary recognition on the natural language description of the equipment anomaly, obtaining a word segmentation sequence. By extracting entities from the word segmentation sequence, a structured entity including equipment identification objects and keywords of abnormal phenomena is obtained. The structured entity is classified by intent. If the core intent is an equipment fault diagnosis task, the equipment to be investigated is matched and located in the equipment asset database based on the equipment identification object, and the preliminary list of anomaly types is obtained by mapping the keywords of abnormal phenomena in the fault feature map.

[0021] In one implementation, the data retrieval engine is activated based on the device to be investigated, and multi-source time-series signal streams are synchronously retrieved from multiple distributed device monitoring modules through a preset interface protocol. The following processing is also performed:

[0022] The data retrieval engine accesses the device asset database based on the unique device identifier of the device to be investigated to retrieve the monitoring data source address list; loads multiple pre-set interface protocol adapters based on the multiple monitoring data source information in the monitoring data source address list to establish network connections with the multiple distributed device monitoring modules; constructs multiple data query requests based on the preliminary anomaly type list and sends them in parallel to the multiple distributed device monitoring modules through the multiple pre-set interface protocol adapters to collect backtracking data and obtain the multi-source time-series signal stream.

[0023] According to a second aspect of the present invention, a system for identifying equipment safety risks based on continuous interaction of mining intelligent agents is provided, the system comprising:

[0024] The system includes the following components: an anomaly description upload unit for equipment maintenance personnel to upload natural language descriptions of equipment anomalies via a mobile terminal text interface; an anomaly parsing and locking unit for the mining agent to receive and parse the natural language descriptions of equipment anomalies, locking the equipment objects to be investigated and a preliminary list of anomaly types; a multi-source time-series retrieval unit for activating a data retrieval engine based on the equipment objects to be investigated, synchronously retrieving multi-source time-series signal streams from multiple distributed equipment monitoring modules through a preset interface protocol; and a state machine fusion analysis unit for the mining agent to initialize a context state machine based on the preliminary list of anomaly types, and then using the context state machine to fuse and analyze the multi-source time-series signal streams. The system outputs an initial fault hypothesis set and an initial fault confidence set; a hypothesis space convergence unit is used to construct a fault hypothesis space convergence strategy for the initial fault hypothesis set based on the confidence deviation of the initial fault confidence set relative to a preset confidence threshold; a compensation signal retrieval unit is used to generate a directional data retrieval instruction based on the fault hypothesis space convergence strategy, driving the data retrieval engine to retrieve multi-source compensation signal streams; and an incremental iterative diagnosis unit is used to input the multi-source compensation signal stream and the multi-source time-series signal stream into the context state machine for incremental iteration of fault root causes until the updated fault confidence set meets the confidence threshold, and outputs a structured equipment fault diagnosis report.

[0025] Beneficial effects of the embodiments of the present invention:

[0026] In the solution provided by this invention, after equipment maintenance personnel upload a natural language description of equipment anomalies via a mobile terminal, the intelligent agent parses the description to identify the equipment to be investigated and a preliminary list of anomaly types. The data retrieval engine is activated to synchronously retrieve multi-source time-series signal streams from the distributed monitoring module through a pre-defined interface protocol. The context state machine is initialized to fuse and analyze the signal streams, outputting an initial set of fault hypotheses and a confidence level set. Based on the confidence deviation, a fault hypothesis space convergence strategy is constructed, generating directional data retrieval instructions to drive the engine to collect multi-source compensation signal streams. The state machine fuses the compensation signals with the original signals for progressive iteration of fault root causes until the updated fault confidence level meets the threshold, at which point a structured diagnostic report containing a fault location matrix and handling strategy is output. This achieves a second-level response time for mining equipment fault diagnosis, automatically generating executable structured reports while significantly shortening the diagnosis cycle, completely eliminating blind spots in manual experience and cross-system data silos, and greatly reducing operation and maintenance costs. Of course, implementing any product or method of this invention does not necessarily require achieving all of the above advantages simultaneously. Attached Figure Description

[0027] 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.

[0028] Figure 1 This invention illustrates a flowchart of a method for identifying equipment safety risks based on continuous interaction of mining intelligent agents.

[0029] Figure 2 A schematic diagram of the structure of the equipment safety risk identification system based on continuous interaction of mining intelligent agents provided by the present invention is shown.

[0030] Figure labeling: 1. Anomaly description upload unit; 2. Anomaly parsing and locking unit; 3. Multi-source timing retrieval unit; 4. State machine fusion analysis unit; 5. Hypothesis space convergence unit; 6. Compensation signal retrieval unit; 7. Progressive iterative diagnosis unit. Detailed Implementation

[0031] To facilitate understanding of the present invention, a more complete description of the invention will be given below with reference to the accompanying drawings, which illustrate preferred embodiments of the invention. However, the invention can be implemented in many different forms and is not limited to the embodiments described herein; rather, these embodiments are provided to enable a more thorough and complete understanding of the disclosure of the invention.

[0032] Furthermore, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0033] In the description of this invention, it should be understood that the terms "center," "longitudinal," "lateral," "length," "width," "thickness," "upper," "lower," "front," "rear," "left," "right," "vertical," "horizontal," "top," "bottom," "inner," "outer," "clockwise," and "counterclockwise," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are only for the convenience of describing this invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this invention.

[0034] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0035] The present invention provides a method and system for identifying equipment safety risks based on continuous interaction of mining intelligent agents, which is used to solve the problems in the prior art of fault diagnosis of mining equipment, such as data silos across monitoring systems and reliance on human experience, which leads to bias in fault diagnosis and delayed feedback of diagnostic results.

[0036] Example 1: See Figure 1 The flowchart of the equipment safety risk identification method based on continuous interaction of mining intelligent agents provided in this embodiment of the invention includes:

[0037] A100: Equipment maintenance personnel can upload natural language descriptions of equipment malfunctions via a mobile terminal text interface.

[0038] Specifically, in this embodiment, equipment maintenance personnel submit natural language descriptions of equipment malfunctions through an industrial mobile terminal text interface that conforms to intrinsically safe mining standards in the underground mining working environment. This terminal preferably adopts a high-contrast anti-glare screen and physical button design, and is optimized for the low-light and high-dust environment of mines.

[0039] The natural language description of the equipment anomaly is a semantic representation of the physical fault of the equipment. It must include key words that are characteristic of the physical fault of the equipment, such as abnormal vibration spectrum, excessive temperature gradient, or distorted current waveform. For example, "abnormal noise of the fan bearing accompanied by excessive temperature rise" implies three core elements: the type of equipment to be diagnosed, the physical anomaly phenomenon, and the operating parameters.

[0040] Subsequently, the mining intelligent agent will initiate a multi-round confidence-driven progressive iterative diagnostic process for equipment failures based on the natural language description of the equipment anomalies.

[0041] A200: The mining intelligent agent receives and parses the natural language description of the equipment anomaly, and identifies the equipment objects to be investigated and a preliminary list of anomaly types.

[0042] In one implementation, the mining agent receives and parses the natural language description of the equipment anomaly, identifies the equipment to be investigated and a preliminary list of anomaly types, and step A200 may further include:

[0043] A210: The mining intelligent agent calls the built-in natural language processing module to perform part-of-speech tagging and entity boundary recognition operations on the natural language description of the equipment anomaly, and obtains a word segmentation sequence.

[0044] A220: By performing entity recognition and extraction on the segmented sequence, a structured entity including device identification objects and abnormal phenomenon keywords is obtained.

[0045] A230: Classify the structured entities by intent. If the core intent is a device fault diagnosis task, then match and lock the device to be investigated in the device asset database based on the device identifier object, and use the abnormal phenomenon keywords to map the fault feature map to obtain the preliminary list of abnormal types.

[0046] Specifically, the aforementioned mine intelligent agent is an intelligent decision-making system specifically designed for diagnosing physical faults in mine equipment. Its built-in natural language processing module is essentially an equipment physical fault description parser. By dynamically identifying the boundaries of three types of keywords in descriptive texts such as abnormal noises from fan bearings and excessively rapid temperature rise of conveyor belts, it generates word segmentation sequences that conform to mine terminology standards.

[0047] The natural language processing module forces the output of structured word segmentation sequences based on the predefined equipment model naming rules, abnormal phenomenon physical characteristics, and operating condition parameter thresholds in the mining equipment body library.

[0048] For example, when the input is "sudden change in the vibration sound of the main ventilator bearing accompanied by excessively rapid temperature rise", the parser (natural language processing module) performs three steps: matching the equipment model keyword with the ventilator model code, mapping the abnormal phenomenon keyword to the bearing wear vibration frequency band characteristics, and converting the operating condition parameter keyword into the temperature sensor monitoring threshold. Finally, it generates a triplet output of the equipment model code, the vibration spectrum abnormal feature vector, and the temperature gradient threshold.

[0049] The mining intelligent agent utilizes named entity recognition technology combined with a mining equipment ontology database to accurately extract equipment identification objects and abnormal phenomenon keywords from a word segmentation sequence. The equipment identification object is a unique physical equipment code registered in the equipment asset database, such as the ventilation fan equipment ID-FJ-2039. The abnormal phenomenon keywords are bound to quantitative fault characteristics such as excessive vibration spectrum amplitude and current waveform distortion rate >15%, forming a structured entity containing both equipment physical attributes and fault characteristics.

[0050] It should be noted that the structured entities strictly follow mining terminology standards to ensure that current waveform distortion is mapped to quantifiable parameters such as excessive current distortion rate and sudden drop in oil pressure is mapped to sudden drop in hydraulic system pressure.

[0051] When classifying the intent of structured entities, the mining intelligent agent uses a fault diagnosis intent classifier to calculate the task probability distribution. This classifier determines the diagnostic intent by analyzing the combination pattern of equipment model and abnormal phenomenon.

[0052] When the equipment type is rotating machinery and the abnormal phenomenon contains vibration spectrum characteristics, the equipment fault diagnosis task is automatically triggered: based on the equipment's unique identifier, the equipment's location coordinates and maintenance records are locked in the equipment asset database; the keywords of the abnormal phenomenon are simultaneously mapped to the fault feature map, and a preliminary list of abnormal types such as bearing wear confidence weight 60% and lubrication failure confidence weight 35% is output; the data call engine is activated to collect the vibration sensor spectrum data and infrared thermal imaging data of the equipment.

[0053] This embodiment obtains diagnostic targets driven by unique equipment identifiers through strong constraint analysis of mine physical fault characteristics and automatic routing of diagnostic intent. This achieves the technical effect of providing quantifiable inputs for subsequent directional retrieval of multi-source time-series signal streams and initialization of context state machines.

[0054] A300: Activate the data call engine according to the device to be investigated, and synchronously retrieve multi-source time-series signal streams from multiple distributed device monitoring modules through a preset interface protocol.

[0055] In one implementation, the data retrieval engine is activated based on the device to be investigated, and multi-source time-series signal streams are synchronously retrieved from multiple distributed device monitoring modules through a preset interface protocol. Step A300 may further include:

[0056] A310: The data retrieval engine accesses the device asset database based on the device's unique identifier to retrieve the monitoring data source address list.

[0057] A320: Load multiple preset interface protocol adapters based on multiple monitoring data source information in the monitoring data source address list, and establish network connections with the multiple distributed device monitoring modules.

[0058] A330: Multiple data query requests will be constructed based on the preliminary list of anomaly types, and sent in parallel to the multiple distributed device monitoring modules through the multiple preset interface protocol adapters to collect running data backtracking, thereby obtaining the multi-source time-series signal stream.

[0059] Specifically, the data retrieval engine accesses the equipment asset database based on the unique identifier of the equipment object to be investigated. The unique identifier is the exclusive code of mining physical equipment such as the ventilation fan ID-FJ-2039.

[0060] The device's unique identifier is used to retrieve the monitoring data source address list from the device asset database. The list contains the physical access addresses of distributed monitoring devices such as vibration sensors and temperature monitors, for example, the IP address and port number of an intrinsically safe vibration sensor used in mining.

[0061] Multiple pre-configured interface protocol adapters are loaded according to the list of monitoring data source addresses. These adapters are communication modules compatible with industrial protocols such as Modbus / TCP and OPC UA, and are used to establish network connections with monitoring modules such as mining vibration spectrum analyzers and infrared thermal imagers.

[0062] Based on a preliminary list of anomaly types, multiple data query requests are constructed. For example, if the anomaly type list includes bearing wear, the request will specify the collection of high-frequency vibration spectrum data in the 200-500Hz range; if it includes lubrication failure, the request will supplement with hydraulic oil temperature gradient data. These requests are sent in parallel to the distributed monitoring module via a pre-configured interface protocol adapter, performing backtracking data collection and ultimately converging into a multi-source time-series signal stream. This signal stream is essentially a time-series collection of physical quantities from the mining equipment, including vibration spectrum waveforms, temperature gradient curves, and current distortion rate sequences.

[0063] This embodiment performs directional acquisition of physical quantities based on the physical fault characteristics of equipment and industrial real-time protocols. Compared with the existing technology that uses manual cross-platform login for decentralized data retrieval, it has the technical advantages of second-level concurrent acquisition and adaptive connection to the mining environment. At the same time, it provides a highly reliable physical signal input effect of binding equipment entities for subsequent context state machine multi-source fusion analysis.

[0064] A400: After the mining agent initializes the context state machine according to the preliminary anomaly type list, it uses the context state machine to fuse and analyze the multi-source time-series signal stream, and outputs the initial fault hypothesis set and the initial fault confidence set.

[0065] In one implementation, after the mining agent initializes the context state machine according to the preliminary anomaly type list, it uses the context state machine to perform fusion analysis on the multi-source time-series signal stream, and outputs an initial fault hypothesis set and an initial fault confidence set. Step A400 may further include:

[0066] A410: Based on the fault feature dimensions of the preliminary anomaly type list, perform directional feature extraction on the multi-source time-series signal stream to construct a preliminary feature vector.

[0067] A420: Create a state machine instance based on the preliminary list of exception types to obtain the context state machine.

[0068] A430: In the context state machine, the preliminary feature vector is used to traverse the structured fault case map to perform similarity matching under multi-source information fusion, thereby obtaining a set of candidate fault modes.

[0069] A440: Perform dynamic confidence decision on the candidate fault mode set to obtain the initial fault hypothesis set and the initial fault confidence set.

[0070] In one implementation, step A400 may further include:

[0071] A411: Perform continuity detection on the multi-source time-series signal stream to locate the sequence of missing data nodes.

[0072] A412: Perform stability assessment on the multi-source time-series signal stream and identify abnormal fluctuation node sequences.

[0073] A413: Based on the missing data node sequence and abnormal fluctuation node sequence, construct a reverse parameter retrieval instruction to obtain the incremental compensation data stream.

[0074] A414: The incremental compensation data stream is superimposed onto the multi-source timing signal stream to obtain a multi-source corrected signal stream.

[0075] A415: Based on the fault feature dimensions of the preliminary anomaly type list, perform directional feature extraction on the multi-source corrected signal stream to construct the preliminary feature vector.

[0076] In one implementation, dynamic confidence decision is performed on the candidate fault mode set to obtain the initial fault hypothesis set and the initial fault confidence set. Step A440 may further include:

[0077] A441: Determine the physical association strength weight coefficient of the fault feature dimension based on the preliminary feature vector, perform weighted similarity calculation on the candidate fault mode set, and generate the initial fault hypothesis set.

[0078] A442: Extract multiple historical case fits, multiple prior probability values ​​and multiple feature matching degrees from the candidate fault mode set, perform dynamic weighted fusion calculation, and output the initial fault confidence set.

[0079] It should be understood that due to communication interruptions or signal interference in the sensor network under complex working conditions in mines, the multi-source time-series signal stream collected in step A300 may have data breakpoints or noise distortion. Therefore, before using this signal stream to extract fault features in this embodiment, it is necessary to perform continuity detection to locate the sequence of missing data nodes.

[0080] Meanwhile, due to sensor measurement drift and electromagnetic interference caused by the high dust environment underground, the multi-source time-series signal stream also has the defect of abnormal amplitude jump. Therefore, this embodiment also detects unexpected violent fluctuation node sequences by analyzing the amplitude change rate and baseline offset of time-series data such as vibration spectrum signal stream and temperature gradient signal stream in the multi-source time-series signal stream.

[0081] For example, in bearing wear diagnosis, when the vibration sensor data shows a sudden increase of 300% in amplitude during a specific period and exceeds the equipment safety threshold, it is marked as an abnormal fluctuation node sequence; if the temperature gradient curve shows an abnormal jump of 50°C in temperature during the lubrication failure analysis interval, it is identified as a noise interference exceeding the limit node.

[0082] For the missing time periods corresponding to the missing data node sequence (e.g., no vibration spectrum data recorded between 08:00 and 08:30) and the abnormal fluctuation intervals corresponding to the abnormal fluctuation node sequence (e.g., a sudden drop in temperature curve at 14:25), a reverse parameter retrieval instruction is generated and distributed to the multiple distributed device monitoring modules to drive the retrospective acquisition of compensation data. This reverse parameter retrieval instruction includes the device's unique identifier, the type of physical quantity to be supplemented, and the time range. For example, for ventilation fan ID-FJ-2039, the original temperature gradient signal for the interval between 14:20 and 14:30 is supplemented.

[0083] The incremental compensation data streams, such as vibration spectrum and temperature gradient, collected through compensation are aligned with timestamps and superimposed onto the original multi-source time-series signal stream to generate a multi-source corrected signal stream that fully covers the diagnostic period. For example, after supplementing the vibration spectrum data from 08:00 to 08:30, the 8-hour vibration waveform for bearing wear analysis becomes continuous and uninterrupted; after correcting the abnormal temperature jump data at 14:25, the temperature gradient curve for lubrication failure diagnosis returns to a smooth change.

[0084] The preliminary anomaly type list is a set of physical fault modes of the equipment (such as bearing wear and lubrication failure) locked based on step A200. This list is derived from the mapping relationship of the fault feature spectrum. For example, bearing wear corresponds to the abnormal features of the vibration spectrum of 200-500Hz. The fault feature dimension covers the direction of physical quantity analysis associated with each anomaly type in the preliminary anomaly type list. For example, bearing wear requires focusing on high-frequency vibration energy, and lubrication failure requires monitoring the temperature rise rate.

[0085] This embodiment selects specific physical quantities from the multi-source correction signal stream based on the fault feature dimension. For example, bearing wear diagnosis only extracts the root mean square value of vibration energy in the 200-500Hz frequency band (e.g., 0.8g²), filtering out irrelevant signals such as temperature gradient; lubrication failure diagnosis only extracts the temperature rise rate per minute (e.g., 5.2℃ / min), ignoring current waveform data, and obtains multi-dimensional fault feature parameters corresponding to multiple preliminary abnormality types in the preliminary abnormality type list, which constitute the preliminary feature vector.

[0086] Conversely, if the integrity verification of the multi-source time-series signal stream passes and the stability fluctuation does not exceed the device safety threshold, then the multi-source time-series signal stream is directly subjected to directional feature extraction based on the fault feature dimension of the preliminary anomaly type list to construct a preliminary feature vector.

[0087] A state machine instance is created based on the preliminary list of anomaly types. Essentially, the state machine instance is a dynamic container for the diagnostic process. Its initialization path is bound to fault characteristic dimensions such as analyzing high-frequency vibration spectra for bearing wear and monitoring temperature rise rates for lubrication failure. For example, in the task of diagnosing bearing wear in a ventilation fan, the state machine instance will use a preset vibration signal analysis as the main path and hydraulic oil temperature monitoring as the auxiliary verification path, forming a dedicated analysis framework for the equipment entity's faults, thus obtaining the context state machine.

[0088] It should be understood that the context state machine is a dynamic analysis container created for the current diagnostic task. For example, after the bearing wear diagnostic task of the fan ID-FJ-2039 is initialized, the state machine continuously tracks the progress of the main path of vibration spectrum analysis and the auxiliary path of lubrication failure verification, forming a diagnostic context bound to the equipment entity.

[0089] The structured fault case map serves as a historical fault knowledge base for mining equipment, storing physical characteristic boundaries such as the vibration energy threshold of bearing wear (0.5–1.0 g²), providing a quantitative matching benchmark for the state machine.

[0090] When the context state machine drives the initial feature vector (such as the vibration energy value of a fan of 0.8g²) to traverse the fault case map, it performs multi-source information fusion matching. For example, the vibration energy value of 0.8g² is used to compare the vibration feature range of 0.5-1.0g² of the bearing wear case in the map, and the physical parameter deviation rate is calculated to generate a similarity of 85%. If the temperature rise rate of 5.2℃ / min is synchronously matched with the temperature rise threshold of the lubrication failure case in the map ≥5℃ / min, then a similarity of 78% is generated. The context state machine fuses the vibration and temperature dual-dimensional matching results and outputs the candidate fault mode set, including but not limited to bearing wear (overall similarity of 85%), lubrication failure (overall similarity of 78%), and other equipment physical fault modes, to realize the progressive linkage analysis of equipment physical fault features and historical cases.

[0091] The mining agent determines the physical correlation strength weight coefficients of each fault feature dimension based on the preliminary feature vector constructed in step A415. The physical correlation strength weight coefficients are predefined by the mining equipment body library and reflect the contribution of different physical quantities to the diagnosis of specific fault modes. For example, in bearing wear diagnosis, the weight coefficient of vibration spectrum energy is 70% (because abnormal vibration is the core characteristic of bearing failure), and the weight coefficient of temperature gradient is 30% (because temperature rise is an auxiliary criterion).

[0092] Based on the physical association strength weight coefficient, a weighted similarity calculation is performed on multiple fault hypotheses in the candidate fault mode set. For example, the original similarity of bearing wear (85%) is multiplied by a weight of 70%, and the similarity of lubrication failure (78%) is multiplied by a weight of 30%, generating a weighted bearing wear hypothesis value of 59.5% and a lubrication failure hypothesis value of 23.4%, which constitute the initial fault hypothesis set. The initial fault hypothesis set is the result of multiple weighted similarity calculations of multiple fault hypotheses, which directly represents the priority ranking of the physical fault modes of the equipment.

[0093] Based on the preliminary feature vector, the physical correlation strength weight coefficient of the fault feature dimension is determined, and the weighted similarity is calculated on the candidate fault mode set to generate the initial fault hypothesis set. This process essentially strengthens the core evidence chain through physical fault mechanisms. For example, the vibration energy value of 0.8g² exceeding the threshold of 0.5g² directly increases the priority of the bearing wear hypothesis.

[0094] Taking the dynamic confidence decision-making process of bearing wear hypothesis in a candidate fault mode set as an example, the mining intelligent agent first extracts the historical case fit rate. This historical case fit rate is achieved by quantifying the deviation rate between the current feature parameters and the historical mean in the structured fault case map. For example, if the vibration energy value of a ventilation fan is 0.8 g², compared to the vibration energy mean of 0.7 g² in historical bearing breakage cases in the map, the calculated physical parameter deviation rate is 14.3%, thus deriving a fit rate of 85.7%. Next, prior probability values ​​are extracted. These prior probability values ​​are derived from the frequency of faults in similar equipment statistically analyzed in the equipment asset database. For example, the bearing fault record frequency for a ventilation fan model is 30% within three years. Finally, feature matching degree is extracted. This feature matching degree measures the temporal correlation between the time-series signal waveform and the fault feature template. For example, the overlap between the vibration spectrum waveform and the bearing wear feature template waveform reaches 90%.

[0095] Based on the predefined dynamic weighting rules of the mining equipment ontology library (historical case consistency weight 50%, prior probability value weight 30%, feature matching degree weight 20%), a weighted fusion calculation is performed. Specifically, 85.7% consistency is multiplied by 50% weight to get 42.85%, 30% prior probability is multiplied by 30% weight to get 9%, and 90% feature matching degree is multiplied by 20% weight to get 18%. The three are added together to output the initial failure confidence of bearing wear of 69.85%, forming a quantitative decision basis for binding equipment entities.

[0096] In this embodiment, the multi-source corrected signal stream obtained through data breakpoint repair and noise distortion calibration is more complete and stable than the original multi-source time-series signal stream. The obtained preliminary feature vector is used as the accurate fault feature input for subsequent state machine fusion analysis.

[0097] A500: Based on the confidence deviation of the initial fault confidence set relative to a preset confidence threshold, construct a fault hypothesis space convergence strategy for the initial fault hypothesis set.

[0098] In one implementation, based on the confidence deviation of the initial fault confidence set relative to a preset confidence threshold, a fault hypothesis space convergence strategy for the initial fault hypothesis set is constructed, and step A500 may further include:

[0099] A510: By sorting the confidence deviations of the initial fault confidence set relative to the confidence threshold in descending order, W low-confidence fault hypotheses are located in the initial fault hypothesis set.

[0100] A520: After matching the W sets of verification indicators for the W low-confidence fault hypotheses in the knowledge base, the missing evidence indicator set is obtained through indicator aggregation, which serves as the convergence strategy for the fault hypothesis space.

[0101] Specifically, taking the initial failure confidence level of 69.85% for the wear of the ventilation fan bearing as an example, the mine intelligent agent sets a confidence threshold of 75% and calculates a confidence deviation of 5.15%. After calculating the confidence deviation of the initial failure confidence set relative to the confidence threshold using the same method, the confidence deviations of all failure hypotheses are sorted in descending order (e.g., the confidence level of lubrication failure is 50% with a deviation of 25%) to locate W low-confidence failure hypotheses. This allows for the identification of weak links that need to be verified first through quantitative deviation. For example, lubrication failure requires special attention because the temperature rise rate of 5.2℃ / min differs from the equipment body library threshold of 5.5℃ / min.

[0102] For the low-confidence hypothesis of lubrication failure, a set of verification indicators, including oil viscosity test values, filter differential pressure data, and oil pump current waveforms, is matched from the knowledge base. This process is repeated for the W low-confidence failure hypotheses, resulting in W sets of verification indicators.

[0103] Because the verification indicator group has cross-system data redundancy and unnecessary verification indicators, redundant indicators are eliminated through indicator aggregation to obtain a missing evidence indicator set. The missing evidence indicator set drives the intelligent agent to initiate a targeted data retrieval command, such as supplementing the collection of oil viscosity parameters for the hydraulic system of the ventilation fan ID-FJ-2039, so as to realize the convergence of the fault hypothesis space from multi-target investigation to single indicator verification.

[0104] A600: Based on the fault hypothesis space convergence strategy, generate a directional data retrieval instruction to drive the data retrieval engine to retrieve multi-source compensation signal streams.

[0105] The essence of retrieving the multi-source compensation signal stream is to trigger precise data backtracking through the gap in physical evidence of the equipment. Taking the low-confidence hypothesis of lubrication failure as an example, the missing evidence index set (oil viscosity detection value) comes from the matching results of the fault mechanism knowledge base. Therefore, the directional data retrieval instruction generated based on this hypothesis is forcibly bound to the equipment entity identifier (ventilator ID-FJ-2039) and the physical quantity type (hydraulic system oil viscosity). The data retrieval engine is driven to retrieve the real-time viscosity time-series signal stream (sampling rate 1Hz) from the hydraulic SCADA system, and simultaneously extract the historical viscosity benchmark value (e.g., 80cP±5%) from the equipment maintenance log, generating a multi-source compensation signal stream containing the real-time curve and the historical benchmark.

[0106] A700: Input the multi-source compensation signal stream and the multi-source timing signal stream into the context state machine to perform progressive iteration of fault root causes until the updated fault confidence set meets the confidence threshold, and output a structured equipment fault diagnosis report.

[0107] In one implementation, the multi-source compensation signal stream and the multi-source timing signal stream are input into the context state machine for progressive iteration of fault root causes until the updated fault confidence set satisfies the confidence threshold, and a structured equipment fault diagnosis report is output. Step A700 may further include:

[0108] A710: Construct the directional data retrieval instruction based on the missing evidence indicator set to drive the data retrieval engine to retrieve multi-source compensation signal streams.

[0109] A720: Based on the fault feature dimensions of the preliminary anomaly type list, perform directional feature extraction on the multi-source compensation signal stream and the multi-source time-series signal stream, and construct an updated feature vector.

[0110] A730: In the context state machine, the updated feature vector is used to traverse the structured fault case map to perform similarity matching under multi-source information fusion, thereby obtaining an updated fault mode set.

[0111] A740: Based on the updated feature vector, perform dynamic weight allocation on the updated fault mode set to generate an updated fault hypothesis set, and then output an updated fault confidence set by performing dynamic confidence decision on the updated fault mode set.

[0112] A750: By analogy, perform incremental iteration of the root cause of the fault until the updated fault confidence set meets the confidence threshold, and output a structured equipment fault diagnosis report.

[0113] Specifically, for the fault characteristic dimensions (temperature rise rate + oil viscosity) of the lubrication failure hypothesis, the mean oil viscosity is extracted from the multi-source compensation signal stream, the temperature rise rate (5.2℃ / min) is extracted from the original multi-source time-series signal stream, and irrelevant signals such as current waveform are filtered out to construct an updated feature vector.

[0114] Using the same data processing procedure as step A400 and its refinement, the context state machine drives the update of feature vectors to traverse the structured fault case map. The oil viscosity of 75 cP matches the viscosity threshold range of 70-80 cP in the lubrication failure case map, generating a similarity of 90%. The temperature rise rate of 5.2℃ / min matches the threshold of ≥5℃ / min, generating a similarity of 78%. The state machine fuses the two-dimensional results to update the fault mode set.

[0115] Based on the updated feature vector, the updated fault mode set is dynamically weighted and an updated fault hypothesis set is generated. Then, by performing dynamic confidence decision on the updated fault mode set, an updated fault confidence set is output.

[0116] The root cause of the fault is iterated incrementally until all confidence data in the updated fault confidence set meets the confidence threshold. A structured equipment fault diagnosis report is then output. The structured equipment fault diagnosis report includes the finally determined fault mode set and the corresponding fault confidence set, as well as the spatial coordinate matrix of the physical fault location of the equipment in the corresponding fault mode set, and the historical fault handling strategies for similar equipment in the fault mode set.

[0117] This embodiment eliminates the delay of manual data retrieval by using a real-time data retrieval engine to provide second-level response to device anomaly descriptions; it automatically repairs data breakpoints and noise distortion, driving precise compensation data acquisition; a multi-round state machine fusion analysis of physical parameters and historical cases, with a confidence-driven closed-loop mechanism that significantly compresses the diagnostic cycle; and finally generates a structured report that integrates accurate fault identification and location with executable handling strategies, completely eliminating blind spots in human experience and cross-system data silos, and significantly reducing the technical effect of mining equipment operation and maintenance costs.

[0118] Example 2: Based on the same inventive concept as the equipment safety risk identification method based on continuous interaction of mining agents in the foregoing examples, this invention provides an equipment safety risk identification system based on continuous interaction of mining agents. See [link to example]. Figure 2 As shown, the system includes:

[0119] The system comprises the following components: Anomaly Description Upload Unit 1, used by equipment maintenance personnel to upload natural language descriptions of equipment anomalies via a mobile terminal text interface; Anomaly Parsing and Locking Unit 2, used by the mining agent to receive and parse the natural language descriptions of equipment anomalies, locking the equipment objects to be investigated and a preliminary list of anomaly types; Multi-Source Timing Retrieval Unit 3, used to activate a data retrieval engine based on the equipment objects to be investigated, synchronously retrieving multi-source timing signal streams from multiple distributed equipment monitoring modules through a pre-set interface protocol; and State Machine Fusion Analysis Unit 4, used by the mining agent to initialize a context state machine based on the preliminary list of anomaly types, and then using the context state machine to perform fusion analysis of the multi-source timing signal streams. The system outputs an initial fault hypothesis set and an initial fault confidence set; a hypothesis space convergence unit 5 is used to construct a fault hypothesis space convergence strategy for the initial fault hypothesis set based on the confidence deviation of the initial fault confidence set relative to a preset confidence threshold; a compensation signal retrieval unit 6 is used to generate a directional data retrieval instruction based on the fault hypothesis space convergence strategy, driving the data retrieval engine to retrieve multi-source compensation signal streams; and an incremental iterative diagnosis unit 7 is used to input the multi-source compensation signal stream and the multi-source time-series signal stream into the context state machine for incremental iteration of fault root causes until the updated fault confidence set meets the confidence threshold, and outputs a structured equipment fault diagnosis report.

[0120] In one implementation, the state machine fusion analysis unit 4 is also used for:

[0121] Based on the fault feature dimensions of the preliminary anomaly type list, directional feature extraction is performed on the multi-source time-series signal stream to construct a preliminary feature vector; a state machine instance is created based on the preliminary anomaly type list to obtain the context state machine; in the context state machine, the preliminary feature vector is used to traverse the structured fault case graph for similarity matching under multi-source information fusion to obtain a candidate fault pattern set; dynamic confidence decision is performed on the candidate fault pattern set to obtain the initial fault hypothesis set and the initial fault confidence set.

[0122] In one implementation, the state machine fusion analysis unit 4 is also used for:

[0123] The multi-source time-series signal stream is subjected to continuity detection to locate the sequence of missing data nodes; the multi-source time-series signal stream is subjected to stability assessment to identify the sequence of abnormal fluctuation nodes; based on the sequence of missing data nodes and the sequence of abnormal fluctuation nodes, a reverse parameter retrieval instruction is constructed to obtain an incremental compensation data stream; the incremental compensation data stream is superimposed on the multi-source time-series signal stream to obtain a multi-source corrected signal stream; according to the fault feature dimension of the preliminary anomaly type list, the multi-source corrected signal stream is subjected to directional feature extraction to construct the preliminary feature vector.

[0124] In one implementation, the state machine fusion analysis unit 4 is also used for:

[0125] Based on the preliminary feature vector, the physical association strength weight coefficient of the fault feature dimension is determined, and the weighted similarity is calculated on the candidate fault pattern set to generate the initial fault hypothesis set. From the candidate fault pattern set, multiple historical case fits, multiple prior probability values ​​and multiple feature matching degrees of multiple fault hypotheses are extracted, and dynamic weighted fusion calculation is performed to output the initial fault confidence set.

[0126] In one implementation, assuming that spatial convergence unit 5 is also used for:

[0127] By sorting the confidence deviations of the initial fault confidence set relative to the confidence threshold in descending order, W low-confidence fault hypotheses are located in the initial fault hypothesis set. After matching the W sets of verification indicators of the W low-confidence fault hypotheses in the knowledge base, the missing evidence indicator set is obtained by indicator aggregation, which serves as the convergence strategy for the fault hypothesis space.

[0128] In one implementation, the progressive iterative diagnostic unit 7 is also used for:

[0129] Based on the missing evidence indicator set, a targeted data retrieval instruction is constructed to drive the data retrieval engine to retrieve multi-source compensation signal streams. According to the fault feature dimensions of the preliminary anomaly type list, targeted feature extraction is performed on the multi-source compensation signal stream and the multi-source time-series signal stream to construct an updated feature vector. In the context state machine, the updated feature vector is used to traverse the structured fault case graph to perform similarity matching under multi-source information fusion, resulting in an updated fault pattern set. Based on the updated feature vector, dynamic weight allocation is performed on the updated fault pattern set to generate an updated fault hypothesis set. Then, dynamic confidence decision-making is performed on the updated fault pattern set to output an updated fault confidence set. This process is repeated iteratively to iterate the root cause of the fault until the updated fault confidence set meets the confidence threshold, at which point a structured equipment fault diagnosis report is output.

[0130] In one implementation, the exception resolution locking unit 2 is also used for:

[0131] The mining agent invokes the built-in natural language processing module to perform part-of-speech tagging and entity boundary recognition on the natural language description of the equipment anomaly, obtaining a word segmentation sequence. By extracting entities from the word segmentation sequence, a structured entity including equipment identification objects and keywords of abnormal phenomena is obtained. The structured entity is classified by intent. If the core intent is an equipment fault diagnosis task, the equipment to be investigated is matched and located in the equipment asset database based on the equipment identification object, and the preliminary list of anomaly types is obtained by mapping the keywords of abnormal phenomena in the fault feature map.

[0132] In one implementation, the exception resolution locking unit 2 is also used for:

[0133] The data retrieval engine accesses the device asset database based on the unique device identifier of the device to be investigated to retrieve the monitoring data source address list; loads multiple pre-set interface protocol adapters based on the multiple monitoring data source information in the monitoring data source address list to establish network connections with the multiple distributed device monitoring modules; constructs multiple data query requests based on the preliminary anomaly type list and sends them in parallel to the multiple distributed device monitoring modules through the multiple pre-set interface protocol adapters to collect backtracking data and obtain the multi-source time-series signal stream.

[0134] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

[0135] The foregoing description of specific exemplary embodiments of the invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of the invention and its practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of the invention, as well as various different choices and variations. The scope of the invention is intended to be defined by the claims and their equivalents.

Claims

1. A method for identifying equipment safety risks based on continuous interaction of mining intelligent agents, characterized in that, include: Equipment maintenance personnel can upload natural language descriptions of equipment malfunctions via a mobile terminal text interface. The mining intelligent agent receives and parses the natural language description of the equipment anomaly, and identifies the equipment objects to be investigated and a preliminary list of anomaly types. The data retrieval engine is activated based on the device to be investigated, and multi-source time-series signal streams are synchronously retrieved from multiple distributed device monitoring modules through a pre-set interface protocol. After the mining agent initializes the context state machine according to the preliminary list of anomaly types, it uses the context state machine to fuse and analyze the multi-source time-series signal stream, and outputs the initial fault hypothesis set and the initial fault confidence set. Based on the confidence deviation of the initial fault confidence set relative to a preset confidence threshold, a fault hypothesis space convergence strategy is constructed for the initial fault hypothesis set. Based on the fault hypothesis space convergence strategy, a directional data retrieval instruction is generated to drive the data retrieval engine to retrieve multi-source compensation signal streams. The multi-source compensation signal stream and the multi-source timing signal stream are input into the context state machine for progressive iteration of fault root causes until the updated fault confidence set meets the confidence threshold, and a structured equipment fault diagnosis report is output.

2. The equipment safety risk identification method based on continuous interaction of mining intelligent agents as described in claim 1, characterized in that, After initializing the context state machine based on the preliminary anomaly type list, the mining agent uses the context state machine to fuse and analyze the multi-source time-series signal stream, outputting an initial fault hypothesis set and an initial fault confidence set, including: Based on the fault feature dimensions of the preliminary anomaly type list, directional feature extraction is performed on the multi-source time-series signal stream to construct a preliminary feature vector; A state machine instance is created based on the preliminary list of exception types to obtain the context state machine; In the context state machine, the preliminary feature vector is used to traverse the structured fault case map to perform similarity matching under multi-source information fusion, thereby obtaining a set of candidate fault patterns; Dynamic confidence decisions are made on the candidate fault mode set to obtain the initial fault hypothesis set and the initial fault confidence set.

3. The equipment safety risk identification method based on continuous interaction of mining intelligent agents as described in claim 2, characterized in that, Also includes: Continuity detection is performed on the multi-source time-series signal stream to locate the sequence of missing data nodes; The stability of the multi-source time-series signal stream is evaluated to identify abnormal fluctuation node sequences; Based on the data missing node sequence and abnormal fluctuation node sequence, a reverse parameter retrieval instruction is constructed to obtain the incremental compensation data stream. The incremental compensation data stream is superimposed onto the multi-source timing signal stream to obtain the multi-source corrected signal stream; Based on the fault feature dimensions of the preliminary anomaly type list, directional feature extraction is performed on the multi-source corrected signal stream to construct the preliminary feature vector.

4. The equipment safety risk identification method based on continuous interaction of mining intelligent agents as described in claim 3, characterized in that, Dynamic confidence decision-making is performed on the candidate fault mode set to obtain the initial fault hypothesis set and the initial fault confidence set, including: Based on the preliminary feature vector, the physical association strength weight coefficient of the fault feature dimension is determined, and the weighted similarity is calculated on the candidate fault mode set to generate the initial fault hypothesis set. Extract the historical case fit, prior probability values ​​and feature matching degree of multiple fault hypotheses from the candidate fault mode set, perform dynamic weighted fusion calculation, and output the initial fault confidence set.

5. The equipment safety risk identification method based on continuous interaction of mining intelligent agents as described in claim 2, characterized in that, Based on the confidence deviation of the initial fault confidence set relative to a preset confidence threshold, a fault hypothesis space convergence strategy is constructed for the initial fault hypothesis set, including: By sorting the confidence deviations of the initial fault confidence set relative to the confidence threshold in descending order, W low-confidence fault hypotheses are located in the initial fault hypothesis set; After matching the W sets of verification indicators for the W low-confidence fault hypotheses in the knowledge base, the missing evidence indicator set is obtained through indicator aggregation, which serves as the convergence strategy for the fault hypothesis space.

6. The equipment safety risk identification method based on continuous interaction of mining intelligent agents as described in claim 5, characterized in that, The multi-source compensation signal stream and the multi-source timing signal stream are input into the context state machine for progressive iteration of fault root causes until the updated fault confidence set satisfies the confidence threshold. A structured equipment fault diagnosis report is then output, including: Based on the missing evidence indicator set, the targeted data retrieval instruction is constructed to drive the data retrieval engine to retrieve multi-source compensation signal streams; Based on the fault feature dimensions of the preliminary anomaly type list, directional feature extraction is performed on the multi-source compensation signal stream and the multi-source time-series signal stream to construct and update the feature vector. In the context state machine, the updated feature vector is used to traverse the structured fault case map to perform similarity matching under multi-source information fusion, thereby obtaining an updated fault pattern set; Based on the updated feature vector, the updated fault mode set is dynamically weighted and an updated fault hypothesis set is generated. Then, by performing dynamic confidence decision on the updated fault mode set, an updated fault confidence set is output. The root cause of the fault is iterated incrementally until the updated fault confidence set meets the confidence threshold, and a structured equipment fault diagnosis report is output.

7. The equipment safety risk identification method based on continuous interaction of mining intelligent agents as described in claim 1, characterized in that, The mining intelligent agent receives and parses the natural language description of the equipment anomaly, identifies the equipment to be investigated and a preliminary list of anomaly types, including: The mining agent invokes the built-in natural language processing module to perform part-of-speech tagging and entity boundary recognition on the natural language description of the equipment anomaly, and obtains a word segmentation sequence; By performing entity recognition and extraction on the segmented sequence, structured entities including device identification objects and abnormal phenomenon keywords are obtained; The structured entities are classified by intent. If the core intent is a device fault diagnosis task, the device to be investigated is matched and located in the device asset database based on the device identifier object. The preliminary list of abnormal types is obtained by mapping the abnormal phenomenon keywords in the fault feature map.

8. The equipment safety risk identification method based on continuous interaction of mining intelligent agents as described in claim 1, characterized in that, Based on the activated data retrieval engine of the device to be investigated, multi-source time-series signal streams are synchronously retrieved from multiple distributed device monitoring modules through a pre-defined interface protocol, including: The data retrieval engine accesses the device asset database based on the unique device identifier of the device to be investigated in order to retrieve the list of monitoring data source addresses; Based on the monitoring data source address list, multiple preset interface protocol adapters are loaded to establish network connections with the multiple distributed device monitoring modules; Multiple data query requests will be constructed based on the preliminary list of anomaly types and sent in parallel to the multiple distributed device monitoring modules through the multiple preset interface protocol adapters to collect operational data backtracking, thereby obtaining the multi-source time-series signal stream.

9. A mine safety risk identification system based on continuous interaction of intelligent agents, characterized in that, For implementing the method steps of any one of claims 1 to 8, comprising: The anomaly description upload unit is used by equipment maintenance personnel to upload natural language descriptions of equipment anomalies through a mobile terminal text interface. An anomaly parsing and locking unit is used by the mining intelligent agent to receive and parse the natural language description of the equipment anomaly, and lock the equipment objects to be investigated and the preliminary list of anomaly types. The multi-source timing retrieval unit is used to activate the data retrieval engine according to the device to be investigated, and synchronously retrieve multi-source timing signal streams from multiple distributed device monitoring modules through a preset interface protocol. The state machine fusion analysis unit is used by the mining intelligent agent to initialize the context state machine according to the preliminary anomaly type list, and then use the context state machine to fuse and analyze the multi-source time-series signal stream to output the initial fault hypothesis set and the initial fault confidence set. The hypothesis space convergence unit is used to construct a fault hypothesis space convergence strategy for the initial fault hypothesis set based on the confidence deviation of the initial fault confidence set relative to a preset confidence threshold. The compensation signal retrieval unit is used to generate a directional data retrieval instruction based on the fault hypothesis space convergence strategy, and drive the data retrieval engine to retrieve multi-source compensation signal streams. The progressive iterative diagnostic unit is used to input the multi-source compensation signal stream and the multi-source timing signal stream into the context state machine to perform progressive iteration of the root cause of the fault until the updated fault confidence set meets the confidence threshold, and output a structured equipment fault diagnosis report.

Images