A key machine pump fault diagnosis system and method

Abstract

The application relates to the field of fault diagnosis, and discloses a key machine pump fault diagnosis system and method, which comprises a task control center and an explosion-proof inspection robot. The task control center comprises a key machine pump acoustic inspection point position database, and the explosion-proof inspection robot comprises a visual recognition module and an acoustic acquisition module. The task control center is used for generating an inspection task according to a preset fixed period, determining a target key machine pump, and acquiring point position information set of the target key machine pump in the key machine pump acoustic inspection point position database. The explosion-proof inspection robot is used for moving to a first target sampling station position based on the point position information set, acquiring a complete set of sampling data set meeting local basic credibility requirements in a low-noise stationary sampling state through the visual recognition module and the acoustic acquisition module, and sending the complete set of sampling data set to the task control center for processing and fault diagnosis. The credibility of a fault determination result is improved.

Description

Technical Field

[0001] This invention relates to the field of fault diagnosis technology, and specifically to a fault diagnosis system and method for key pumps. Background Technology

[0002] Critical pumps and machinery in hazardous chemical plants are typically located in environments characterized by high temperature, high humidity, high dust levels, high noise levels, corrosive gases, dense pipelines, and restricted personnel access. Leaks, blockages, cavitation, bearing damage, mechanical seal malfunctions, or motor abnormalities in these pumps and machinery can lead to media leaks, equipment fluctuations, systemic shutdowns, fires, explosions, or personal injury risks. Therefore, high-frequency inspections and condition diagnostics of critical pumps and machinery are necessary through unmanned, remote, and data-driven methods.

[0003] Current methods for managing critical pumps and machinery mainly include manual on-site inspections, online monitoring using fixed sensors, and routine inspections using robots. While these methods can detect some obvious anomalies, they still have the following shortcomings in the complex environments of hazardous chemical plants.

[0004] 1. Manual inspection requires personnel to enter the plant area, where there may be toxic and harmful gases, corrosive media, high-temperature equipment, flammable and explosive atmospheres, and noisy environments, posing a high risk of personnel exposure.

[0005] 2. Manual listening, temperature measurement, handheld vibration measurement, and experience-based judgment rely on the capabilities of the inspection personnel. Different personnel have different measurement positions, angles, recording methods, and abnormal judgment standards, resulting in poor data comparability.

[0006] 3. Manual inspections are usually performed on a fixed schedule, which makes it difficult to cover the entire continuous operation of the equipment. When rapid deterioration occurs in the equipment between two inspection intervals, it may not be detected in time.

[0007] 4. Fixed sensor deployment is costly, and hazardous chemical sites have limitations in terms of explosion protection, corrosion protection, wiring, maintenance, and installation space, making it unsuitable for large-scale deployment of contact sensors in all critical locations.

[0008] 5. Ordinary inspection robots typically rely on cameras, infrared thermometry, meter recognition, or gas detection, and have limited ability to identify changes in mechanical condition such as early bearing damage, coupling asymmetry, cavitation, and abnormal lubrication.

[0009] 6. Diagnostic algorithms based solely on audio or vibration features are easily affected by crosstalk from adjacent pumps, pipe reflections, environmental noise, and robot noise in hazardous chemical installation areas. If the sampling location, sampling distance, and sampling posture are inconsistent, false alarms or missed alarms are likely to occur. Summary of the Invention

[0010] This invention provides a critical pump fault diagnosis system and method to address the shortcomings of existing technologies, which mainly include manual on-site inspection, fixed sensor online monitoring, and ordinary inspection robot inspection. Although these methods can detect some obvious anomalies, they are still insufficient in the complex environments of hazardous chemical enterprises.

[0011] In a first aspect, the present invention provides a critical pump fault diagnosis system, the system comprising: The system comprises a mission control center and an explosion-proof inspection robot. The mission control center includes a database of acoustic inspection points for key pumps and machinery, while the explosion-proof inspection robot includes a visual recognition module and an acoustic acquisition module. The mission control center generates inspection tasks according to a preset fixed cycle, identifies target key pumps and machinery based on these tasks, and retrieves the location information set of the target key pumps and machinery from the acoustic inspection point database. It then sends this location information set and the database to the explosion-proof inspection robot. The explosion-proof inspection robot, when moving to the first target sampling station based on the location information set, uses its visual recognition module to... The module and acoustic acquisition module acquire a complete set of sampling datasets that meet the local basic reliability requirements under low-noise parking sampling conditions, and send the complete set of sampling datasets to the task control center. The low-noise parking sampling conditions meet multi-dimensional joint gating conditions. The complete set of sampling datasets includes background sound field signals, robot body noise templates, acoustic signals of target key pumps, and auxiliary state datasets. The task control center is also used to process the complete set of sampling datasets sequentially, including data processing, sound field compensation, adaptive noise suppression, multi-station sound source attribution discrimination, diagnostic calculations, and working condition baseline matching, and obtain the target fault diagnosis results of the target key pumps.

[0012] The critical pump fault diagnosis system provided by this invention automates equipment inspection by periodically generating and distributing inspection point information sets through a task control center. This eliminates the need for manual on-site scheduling and unifies the sampling benchmarks for all pumps, ensuring horizontal and vertical comparability of multi-round inspection data. Furthermore, after the explosion-proof inspection robot moves to the sampling station, it collects a complete set of sampling data and performs pre-processing local basic reliability verification. This filters out invalid sampling data caused by robot shaking, posture deviation, and excessive body noise, reducing bandwidth consumption for invalid data transmission and lowering the load of repetitive backend calculations. Simultaneously, by limiting low-noise parking and multi-dimensional joint gating, the system unifies the environmental and equipment posture constraints for each acoustic acquisition, eliminating acoustic and vibration signal interference caused by differences in sampling posture and equipment operating status, significantly reducing the probability of false alarms and missed alarms in fault diagnosis. Finally, by sequentially executing data processing, sound field compensation, adaptive noise suppression, multi-station sound source attribution determination, and operating condition baseline matching diagnosis, the task control center can layer-by-layer isolate four types of interference: crosstalk from adjacent equipment, pipeline reflection, robot self-noise, and operating condition fluctuations. This allows for precise identification of abnormal sound sources belonging to the target pump, improving the sensitivity of early-stage minor mechanical fault identification. Therefore, by implementing this invention, through the collaborative mechanism of robot-based pre-inspection quality control and multi-layered noise reduction and source tracing diagnosis in the background, the industry pain points of high safety risks during manual inspections in hazardous chemical plants, high deployment costs of fixed sensors, and the inability of ordinary robots to identify early acoustic and vibration faults in pumps are solved. This achieves stable output of highly reliable fault judgment results without deploying a large number of contact sensors.

[0013] In one optional implementation, the explosion-proof inspection robot further includes: a navigation path planning module, a movement control module, and a local control unit; the navigation path planning module is used to plan a path based on the point information set to obtain a target movement path, and send the target movement path to the movement control module through the local control unit; the movement control module is used to control the movement of the explosion-proof inspection robot based on the target movement path, and when the explosion-proof inspection robot moves to the first target sampling station, it sends a movement completion command to the local control unit, so that the local control unit sends a visual acquisition command to the visual recognition module based on the movement completion command.

[0014] The critical pump fault diagnosis system provided by this invention, combined with a set of location information, plans a safe movement path through a navigation path planning module. It can automatically avoid hazardous areas and obstacles within the plant area, balancing traffic safety with the accessibility of sampling points, eliminating the need for manual route planning for each pump. Furthermore, the system relays path commands through a local control unit and triggers visual acquisition after the explosion-proof inspection robot moves to the first target sampling station. Local interaction within the robot's internal modules eliminates the need for real-time communication with the backend, reducing communication latency and enabling complete pre-sampling identification even in weak network environments. Therefore, by implementing this invention, autonomous and safe robot movement and local process scheduling are achieved. It can adapt to the unstable communication and complex area divisions of hazardous chemical plants, reducing the pressure on backend scheduling and improving the continuity and safety of inspection operations.

[0015] In one optional implementation, the visual recognition module is used to acquire a corresponding target area image based on a visual acquisition command, recognize the target area image to obtain a visual recognition result, and send the visual recognition result to the local control unit; the local control unit is used to determine whether the detection object contained in the target area image is a target key pump based on the visual recognition result and the key pump acoustic inspection point database, and when the detection object is a target key pump, to control the explosion-proof inspection robot to be in a low-noise parking sampling state based on preset parking conditions, and to send an acoustic acquisition command to the acoustic acquisition module; the acoustic acquisition module, This system is used to collect background sound field signals, robot body noise templates, acoustic signals of key target pumps, and auxiliary state datasets based on acoustic acquisition commands, and then send these data to the local control unit. The local control unit is also used to encapsulate the background sound field signals, robot body noise templates, acoustic signals, and auxiliary state datasets to obtain a complete set of sampled datasets, perform local basic credibility judgment on the complete set of sampled datasets, and send the complete set of sampled datasets to the task control center when the local basic credibility meets a first preset threshold requirement.

[0016] The critical pump fault diagnosis system provided by this invention, by combining visual recognition results and a point database to verify the identity of the target pump, can prevent the robot from mistakenly sampling similar surrounding equipment, thus eliminating diagnostic errors caused by cross-device data confusion. Furthermore, after successful identity verification, a pause and simultaneous acquisition of a complete set of acoustic data are triggered, simultaneously storing three types of reference samples: target sound source, environmental noise, and robot self-noise, providing a complete reference data source for subsequent layered noise reduction compensation. Furthermore, by using a local control unit to perform local basic reliability judgment on the collected data before sending it to the task control center, a large amount of low-quality raw acoustic data backhaul is reduced, lowering the data storage and transmission pressure on edge and backend servers. Therefore, by implementing this invention, through establishing a dual quality control mechanism of pre-sampling device identity verification and local initial screening of sampled data, the problems of incorrect and invalid sampling are avoided from the source, while simultaneously optimizing the overall data transmission and storage resource consumption of the system.

[0017] In one optional implementation, the explosion-proof inspection robot further includes: a sampling posture adjustment mechanism; a local control unit, which is further configured to send a first adjustment command to the sampling posture adjustment mechanism when the object being inspected is not the target critical pump; the sampling posture adjustment mechanism is configured to adjust the position and orientation of the explosion-proof inspection robot based on the first adjustment command, and send a first adjustment completion command to the local control unit according to the adjusted position and orientation, so that the local control unit sends a new visual acquisition command to the visual recognition module based on the first adjustment completion command.

[0018] The critical pump fault diagnosis system provided by this invention adjusts the position and orientation of the explosion-proof inspection robot through a sampling posture adjustment mechanism and re-acquires visual data when the object of inspection is not the target critical pump. This achieves the goal of correcting the sampling perspective without manual remote intervention. Simultaneously, it improves the first-pass rate of equipment identification by autonomously correcting robot orientation and distance deviations. Therefore, by implementing this invention, the robot's autonomous posture correction and re-identification capability can adapt to complex working conditions with dense equipment layouts and obstructed visual markers in the factory, improving the automation level of inspections and reducing the frequency of manual remote verification.

[0019] In one optional implementation, the local control unit is further configured to send a second adjustment instruction to the sampling posture adjustment mechanism when the local basic confidence level does not meet the first preset threshold requirement; the sampling posture adjustment mechanism is further configured to adjust the position and orientation of the explosion-proof inspection robot based on the second adjustment instruction, and send a second adjustment completion instruction to the local control unit according to the adjusted position and orientation, so that the local control unit sends a new acoustic acquisition instruction to the acoustic acquisition module based on the second adjustment completion instruction.

[0020] The critical pump fault diagnosis system provided by this invention adjusts the position and orientation of the explosion-proof inspection robot and re-acoustic sampling through a sampling posture adjustment mechanism when the local basic reliability does not meet the first preset threshold requirement. This optimizes the sampling angle and distance within the current optimal sampling station, improving the sampling signal-to-noise ratio as much as possible without moving the robot and shortening the time required for a single inspection. Therefore, by implementing this invention, the adaptive resampling mechanism within the station balances sampling data quality and inspection efficiency, avoiding redundancy in the inspection process caused by directly switching to a backup station due to slight posture deviations.

[0021] In one optional implementation, the local control unit is further configured to send a movement command to the movement control module when the local basic trust level does not meet the first preset threshold requirement, so that the movement control module controls the explosion-proof inspection robot to move to the second target sampling station or the backup sampling station based on the movement command.

[0022] The critical pump fault diagnosis system provided by this invention, when the local basic reliability does not meet the first preset threshold requirement, controls the explosion-proof inspection robot to move to the second target sampling station or the backup sampling station for resampling through the mobile control module. This forms a two-level resampling fault tolerance mechanism of adjusting attitude in place and changing station, which can adapt to local high noise and severely obstructed points in the plant area and ensure that effective acoustic and vibration data can be collected for each pump.

[0023] In one optional implementation, the task control center includes: a data analysis engine and a diagnostic result database; the data analysis engine is used to perform a comprehensive confidence assessment of the entire set of sampled datasets across all dimensions, and when the comprehensive confidence assessment across all dimensions meets the second preset threshold requirement, it sequentially performs data processing, sound field compensation, adaptive noise suppression, multi-station sound source attribution discrimination, diagnostic calculation, and operating condition baseline matching on the entire set of sampled datasets, and obtains the target fault diagnosis results of the target key pump; the diagnostic result database is used to store the target fault diagnosis results.

[0024] The critical pump fault diagnosis system provided by this invention uses a data analysis engine to perform comprehensive reliability assessment of the entire sampled dataset from all dimensions, avoiding the misfiltering of qualified data caused by relying solely on simple local gating by the robot. Furthermore, by storing the diagnostic results in a database, a full-cycle equipment health data archive can be formed, which helps support fault trend retrospection and horizontal comparative analysis of similar equipment. Therefore, by implementing this invention, a two-layer reliability verification system of robot-based coarse judgment and background global fine judgment is constructed, while simultaneously completing the full lifecycle archiving of equipment health data, providing complete data support for long-term fault trend analysis and equipment operation and maintenance optimization.

[0025] In one optional implementation, the task control center is further configured to generate anomaly review tasks based on fixed sensor alarms, DCS operating condition anomalies, manual remote commands, or historical trend anomalies, and to determine target key pumps based on the anomaly review tasks; the task control center is further configured to acquire operating conditions where historical acoustic and vibration characteristics continuously deviate from the baseline, and to generate fault trend tracking and diagnosis tasks based on the operating conditions.

[0026] The critical pump fault diagnosis system provided by this invention generates anomaly review tasks based on fixed sensor alarms, DCS operating condition anomalies, manual remote commands, or historical trend anomalies. This breaks through the single mode of scheduled inspections, enabling immediate targeted re-inspections of the process system when anomalies occur, thus shortening the response time to sudden faults. Furthermore, by generating fault trend tracking diagnosis tasks based on operating conditions that continuously deviate from the baseline according to historical acoustic and vibration characteristics, it can proactively capture early, latent faults that deteriorate slowly, compensating for the shortcomings of periodic inspections in continuously tracking gradual equipment damage. Therefore, by implementing this invention, the coordinated scheduling of three types of tasks—scheduled inspections, emergency anomaly review, and long-term trend tracking—is achieved, taking into account both routine equipment health management and proactive early warning of sudden and gradual faults, thereby improving the completeness of system fault coverage.

[0027] In a second aspect, the present invention provides a method for diagnosing faults in critical pumps, used in the task control center of a critical pump fault diagnosis system described in the first aspect or any corresponding embodiment thereof; the method includes: The system receives a complete set of sampling datasets of the target critical pump from the explosion-proof inspection robot in the critical pump fault diagnosis system. It then processes the datasets, performs sound field compensation, adaptive noise suppression, and multi-station sound source attribution to obtain a compensated acoustic-vibration response feature set. Indices are extracted from the acoustic-vibration response feature set, and an acoustic-vibration feature vector set is constructed. Based on the acoustic inspection point database of the critical pump, a real-time operating condition vector is constructed, and a comprehensive acoustic-vibration anomaly index is built based on the real-time operating condition vector and the acoustic-vibration feature vector. The basic state level of the target critical pump is determined based on the comprehensive acoustic-vibration anomaly index. Finally, based on the comprehensive acoustic-vibration anomaly index, the acoustic-vibration feature vector set, the real-time operating condition vector, and the basic state level, fault diagnosis is performed using preset fault type acoustic-vibration judgment rules to obtain the target fault diagnosis result for the target critical pump.

[0028] The key pump fault diagnosis method provided by this invention, through hierarchical processing of acoustic data and construction of a multi-dimensional acoustic vibration feature vector, can completely preserve the differentiated acoustic vibration features corresponding to the fault, such as impact, frequency band, and envelope, avoiding the loss of fault features caused by a single indicator. Furthermore, by constructing a real-time operating condition vector and matching it with a historical baseline, the influence of operating condition fluctuations such as speed, pressure, and flow on the acoustic vibration signal can be offset, eliminating false alarms caused by changes in normal operating conditions. Finally, by comprehensively determining the equipment level based on anomaly indices and combining multi-dimensional features for joint fault diagnosis, the degree to which the equipment deviates from its healthy state is quantified, while simultaneously achieving accurate differentiation of different types of pump faults such as bearing, cavitation, and misalignment. Therefore, by implementing this invention, and by establishing an operating condition-adaptive multi-dimensional acoustic vibration fault quantitative diagnosis logic, the shortcomings of traditional fixed threshold diagnosis in adapting to pumps with varying operating conditions are solved, simultaneously achieving accurate identification of equipment health classification and fault types.

[0029] In one optional implementation, a comprehensive acoustic-vibration anomaly index is constructed based on real-time operating condition vectors and acoustic-vibration feature vectors, including: Based on real-time operating condition vectors, the mean and covariance matrix of the normal acoustic and vibration baseline matching the target key pumps are determined; based on the acoustic and vibration feature vector set, the mean and covariance matrix of the normal acoustic and vibration baseline are determined; and based on the normalized abnormal deviation of the operating condition, a comprehensive acoustic and vibration anomaly index is constructed.

[0030] The critical pump fault diagnosis method provided by this invention determines the mean and covariance matrix of normal acoustic and vibration baselines that match the operating conditions of the target critical pump by combining real-time operating condition vectors. This enables the construction of a dynamic health reference standard using historical normal samples, rather than a fixed static threshold, thus adapting to performance changes throughout the equipment's lifecycle. Furthermore, by calculating the normalized abnormal deviation of operating conditions and generating a weighted comprehensive acoustic and vibration anomaly index, the method quantifies the degree of difference between the current equipment's health status and that of equipment under the same operating conditions. This allows the index value to intuitively reflect the severity of the fault and the rate of degradation. Therefore, by implementing this invention, the interference of individual equipment differences and differences in operating conditions on the diagnostic results is reduced, thereby enabling the output of a quantitative indicator of equipment health with a unified measurement standard.

[0031] In an optional implementation, the method further includes: generating a hierarchical closed-loop handling plan based on the target fault diagnosis results.

[0032] The critical pump fault diagnosis method provided by this invention, by combining diagnostic grading with differentiated handling solutions, can automatically verify, remotely alarm, and issue maintenance work orders based on fault risk matching, achieving a fully automated closed-loop process for fault detection and handling. Therefore, by implementing this invention, the process link between fault diagnosis and on-site maintenance is streamlined, forming a tiered handling closed loop, reducing the workload of manual judgment and work order dispatching for maintenance personnel, and improving the response efficiency for hazardous chemical pump fault handling. Attached Figure Description

[0033] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0034] Figure 1 This is a structural block diagram of a key pump fault diagnosis system according to an embodiment of the present invention; Figure 2 This is a flowchart illustrating a key pump fault diagnosis method according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the overall scheme of the key pump acoustic vibration condition monitoring and intelligent diagnosis system according to an embodiment of the present invention; Figure 4 This is a flowchart illustrating the key pump acoustic and vibration condition monitoring and intelligent diagnosis method according to an embodiment of the present invention; Figure 5 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0035] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0036] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0037] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0038] Key pumps and compressors are subject to factors such as rotor rotation, fluid impact, pipeline load, foundation constraints, changes in medium condition, and fluctuations in process conditions during long-term operation. Their failures typically exhibit both gradual and sudden characteristics. Early failures may manifest only as subtle changes in operating sound, increased localized impact, or frequency band energy variations; as the failure progresses, significant vibration, temperature rise, leakage, pressure fluctuations, or abnormal flow may occur. (See Table 1 below.)

[0039] Table 1. Typical Failure Manifestations of Key Pumps and Machinery

[0040] This invention provides a critical pump fault diagnosis system. Through a collaborative mechanism of robot-based pre-inspection and multi-layer noise reduction and source tracing diagnosis in the background, it solves the industry pain points of high safety risks of manual inspection in hazardous chemical plants, high deployment costs of fixed sensors, and the inability of ordinary robots to identify early acoustic and vibration faults of pumps. It achieves stable output of highly reliable fault judgment results without deploying a large number of contact sensors.

[0041] As an optional application scenario of this invention, a critical pump fault diagnosis system is provided, such as... Figure 1 As shown, the critical pump fault diagnosis system 1 includes a task control center 2 and an explosion-proof inspection robot 3. Furthermore, the task control center 2 includes a critical pump acoustic inspection point database 21; the explosion-proof inspection robot 3 includes a visual recognition module 31 and an acoustic acquisition module 32.

[0042] Optionally, the task control center 2 is used to generate inspection tasks according to a preset fixed cycle, and based on the inspection tasks, determine the target key pumps, and obtain the point information set of the target key pumps from the key pump acoustic inspection point database, and send the point information set and the key pump acoustic inspection point database to the explosion-proof inspection robot.

[0043] In one optional embodiment, the preset fixed cycle represents the time rule for periodic automatic triggering of pump inspections, pre-configured by maintenance personnel in the task control center, which can adapt to the continuous and uninterrupted operation of critical pumps in hazardous chemical enterprises. Furthermore, the cycle can be customized by hour / shift / day / week. For example, short-cycle, high-frequency inspections can be configured for high-risk moving equipment such as media transfer pumps and reaction feed pumps; long-cycle inspections can be configured for low-risk auxiliary pumps such as circulating pumps and cooling pumps.

[0044] Furthermore, the inspection task refers to the standardized operation instruction package generated and issued by the task control center 2 to the explosion-proof inspection robot 3, which can be divided into three categories: routine timed inspection tasks, anomaly review tasks, and fault trend tracking tasks. Furthermore, this inspection task encapsulates constraints such as a list of target pumps to be inspected, restricted / prohibited areas within the plant, recommended and backup sampling stations, sampling posture parameters, sampling duration, data upload requirements, and re-sampling trigger thresholds.

[0045] Furthermore, the target key pumps refer to the continuous / high-frequency start-stop equipment used in the production, storage and transportation of flammable, explosive, toxic and corrosive media in hazardous chemical enterprises. These are the designated inspection targets for this inspection task, specifically including media transfer pumps, circulation pumps, cooling pumps, reaction feed pumps, reflux pumps, compressor-supporting pump sets, and pumps related to core process logistics.

[0046] Furthermore, the key pump acoustic inspection point database not only records the pump's position coordinates but also information related to acoustic sampling, enabling the robot to perform sampling according to stable and repeatable positions and postures. Specifically, this can include the following: 1. Key pump and machine serial numbers, names, equipment types, affiliated units, media types, and risk levels; 2. Location coordinates of pumps and motors on the plant area map, accessible routes, hazard level, and obstacle avoidance information; 3. Recommended sampling station location, backup sampling station location, target sampling direction, sampling height, and sampling distance range for the robot; 4. Target detection area information, such as pump body, bearing housing, coupling area, mechanical seal area, inlet pipe section, outlet pipe section, and valve group area; 5. Visual identification information, such as QR codes, equipment nameplates, outline features, pipeline interface locations, or preset visual markers; 6. Historical acoustic baseline data, and operating condition baselines corresponding to speed, flow rate, pressure, temperature, load, valve position, or operating mode; 7. Location information of adjacent pumps, pipelines, valves, compressed air, ventilation equipment, etc., which may form sources of interference.

[0047] Furthermore, the location information set may include the location coordinates of the target key pump, recommended sampling station, backup sampling station, target sampling direction, sampling distance range, historical acoustic baseline, applicable working condition information, hazardous area information, on-site obstacles, and the current position of the robot.

[0048] In one optional embodiment, the task control center 2 automatically generates standardized inspection operation instructions for key pumps and machines throughout the plant by reading the pre-configured timing cycle parameters of operation and maintenance, without the need for manual initiation of a single inspection.

[0049] For example, maintenance personnel can configure tiered cycle rules in the task control center in advance. Specifically, the inspection frequency is divided according to the risk of pump media and operating load, i.e., short cycles are set for high-risk feed pumps and return pumps, and long cycles are set for ordinary cooling circulation pumps.

[0050] Furthermore, the task control center 2 is equipped with a timing and scheduling module (not shown in the figure), which automatically triggers the task generation logic when a preset periodic node is reached. Then, it retrieves all equipment ledgers from the key pump acoustic inspection point database 21 and splits the inspection sub-tasks into groups according to the plant area, pipe gallery, and storage tank area to avoid long-distance round trips for the robot across areas and reduce inspection time.

[0051] Furthermore, when an inspection task is generated, corresponding constraint parameters are bound simultaneously, which may include the recommended sampling station for the corresponding pump, the backup sampling station, the sampling direction, the sampling distance, the parking gate control conditions, the first threshold of local credibility, and the second threshold of comprehensive credibility in the background.

[0052] Furthermore, tasks can be prioritized. For example, tasks involving high-explosive or highly toxic media pumps within the same cycle can be marked as high priority, allowing the explosion-proof inspection robot 3 to execute them first. Furthermore, after the above operations are completed, an inspection task data package containing a unique task number, execution time, abnormal re-sampling trigger rules, and data storage archiving requirements can be generated.

[0053] In one optional embodiment, the task control center 2 parses the inspection task data packet, filters out all pump equipment that needs to complete acoustic and vibration monitoring in this cycle, and then locks the unique identifier of each key pump to be tested.

[0054] For example, the task control center 2 parses the equipment number list within the inspection task and matches it with the equipment files in the key pump acoustic inspection point database 21 to verify the current operating status of the equipment. Among them, stopped equipment is automatically removed, and only continuously running, high-frequency pump start-up equipment is retained.

[0055] Furthermore, the basic information of the equipment in the associated location database can include the pump name, the production unit to which it belongs, the medium type, the safety risk level, and the distribution of adjacent interference noise sources. At the same time, it matches the plant area navigation grid map and marks the coordinates of each target pump, the surrounding restricted areas, obstacles, and access routes.

[0056] Furthermore, the paths of multiple target pumps are optimized and sorted, and the order of detection is adjusted according to the principle of shortest travel distance and least crossing of high-risk areas, so as to reduce robot power consumption and travel time in high-risk areas of the factory. In this way, a complete list of key target pumps for this task can be output, and the relevant sampling parameters of each device can be bound.

[0057] In an optional embodiment, for each target pump in the list, the corresponding sampling, navigation, identification, and working condition parameters can be extracted from the key pump acoustic inspection point library 21 and packaged into an explosion-proof inspection robot 3 under an independent point information set.

[0058] For example, the location database is retrieved based on the unique identifier of the target pump, and the corresponding stored information is extracted layer by layer: ①Basic equipment information: pump number, equipment type, unit to which it belongs, medium, and risk level; ② Factory area spatial navigation information: coordinates, passable paths, hazard level, obstacle distribution, robot power / communication constraints; ③ Sampling station parameters: recommended sampling station, backup sampling station, sampling height, and standard sampling distance range; ④ Parameters of the detection area: key acoustic and vibration collection areas such as bearing housings, couplings, mechanical seals, inlet and outlet pipelines, and valve groups; ⑤ Visual recognition matching parameters: equipment QR code, nameplate text, pump body outline, pipeline interface, preset visual markings; ⑥ Operating condition baseline data: Historical standard acoustic baselines corresponding to different speeds, flow rates, pressures, temperatures, valve positions, and loads; ⑦ Data on interference sources: Location of nearby pumps, fans, valves, exhaust systems, and compressed air equipment.

[0059] Furthermore, by removing irrelevant equipment data from the location database that are not for the target pump, retaining only the parameters of the current detection object, and then supplementing auxiliary parameters such as the sampling attitude adjustment cost, path hazard penalty coefficient, and communication quality threshold required for path planning, a location information set corresponding to a single target pump can be finally generated.

[0060] Furthermore, if there are multiple devices, multiple independent location information sets are generated and packaged together.

[0061] Furthermore, the mission control center 2 sends a set of location information to the explosion-proof inspection robot 3 in the plant area, while simultaneously synchronizing the basic index data of the location database to support the local autonomous navigation, visual verification, and sampling posture adjustment of the explosion-proof inspection robot 3. Additionally, the original files, full historical baselines, and diagnostic models of the complete acoustic inspection location database 21 for key pumps are permanently stored in the mission control center 2 / edge server. The explosion-proof inspection robot 3 only caches temporary location parameters and does not persistently store the full location database.

[0062] Optionally, the explosion-proof inspection robot 3 is used to move to the first target sampling station based on the point information set, and based on the key pump acoustic inspection point library, to obtain a complete set of sampling data that meets the local basic credibility requirements under low-noise parking sampling state through the visual recognition module 31 and the acoustic acquisition module 32, and send the complete set of sampling data to the task control center 2.

[0063] In one optional embodiment, the first target sampling station point represents the highest priority recommended sampling station point stored in the key pump acoustic inspection point point library 21, which is a standard point point preset by the system and given priority for the robot to dock and carry out acoustic collection.

[0064] Furthermore, the low-noise parking sampling state indicates that the explosion-proof inspection robot enters a stable sampling operation mode that meets multi-dimensional joint gating conditions after arriving at the sampling station and completing the identity verification of the target pump. In this mode, all moving mechanisms that generate additional noise and vibration are shut down / unloaded, eliminating interference from the robot's own motion and ensuring that the acoustic acquisition module collects pure and stable acoustic and vibration data. This is a necessary prerequisite for collecting effective diagnostic data.

[0065] Among them, the multi-dimensional joint gating condition represents a set of parallel constraint judgment criteria for determining whether the robot has reached an effective low-noise parking sampling state. That is, acoustic acquisition is allowed to start only when all conditions are met simultaneously. If any one condition is not met, the parking is deemed invalid and the robot needs to adjust its posture or change its position. It can include five judgment dimensions: 1. Chassis motion constraints: The robot chassis speed is 0, with no movement or sliding vibration; 2. Sampling posture constraints: The deviation angle of the acoustic acquisition module from the standard sampling direction is less than the preset threshold, and the sampling distance falls within the specified range of the point database; 3. Body noise constraint: The robot's own operating noise energy is lower than the preset noise threshold; 4. Vehicle stability constraints: The pitch and roll of the vehicle body are within the allowable range; 5. Visual recognition constraint: The visual recognition module identifies the target pump with a matching confidence level that meets the minimum standard.

[0066] Furthermore, the complete set of sampling datasets may include background sound field signals, robot body noise templates, acoustic signals of target key pumps, and auxiliary state datasets.

[0067] Among them, the background sound field signal represents the environmental interference acoustic data collected by the robot synchronously before and after sampling in the target sampling area. It is used to characterize all non-target sound sources around the target pump, which may include the operating sound of adjacent pumps, the sound of fluid reflection in pipelines, the noise of exhaust equipment, the sound of compressed air flow, the noise of valve opening and closing, and other ambient noises.

[0068] Furthermore, the robot body noise template represents the self-noise sample collected by the robot in a stationary, unloaded state when the acoustic acquisition module is not facing the target pump. This sample serves as standardized reference data for the robot's inherent interference noise sources. Furthermore, this noise source can include vibrations from the robot's chassis motors, cooling fans, power modules, gimbal / robotic arm structural resonance, and other components.

[0069] Furthermore, the acoustic signals of the target key pumps represent the original operating sound and vibration signals collected by the acoustic acquisition module, which is oriented towards the pump body, bearing housing, coupling, mechanical seal, and other fault detection parts of the target pump. These signals are used to reflect the mechanical health status of the pump and may include basic rotor operating sound, periodic bearing impact sound, random high-frequency noise from cavitation, seal friction sound, and fluid load sound in the pipeline.

[0070] Furthermore, the auxiliary state dataset represents multi-dimensional supporting operating condition, environmental, and visual sensor data acquired synchronously with the acoustic signal. It is used to assist in baseline matching of operating conditions and cross-verification of faults. While not directly used as the basis for acoustic and vibration diagnosis, it can significantly improve diagnostic accuracy. Specifically, it can include four types of data: 1. Visual image data: Real-time images of pump and motor appearance, pipelines, and valve assembly; 2. Infrared temperature measurement data: temperature of pump body, bearings, and mechanical seal components; 3. Environmental monitoring data: Concentration of toxic and flammable gases on site; 4. Process operating data: real-time speed of pumps and motors, medium flow rate, inlet and outlet pressure, medium temperature, valve opening, and equipment operating load.

[0071] In an optional embodiment, the explosion-proof inspection robot 3 further includes a navigation path planning module 33, a movement control module 34, a local control unit 35, and a sampling posture adjustment mechanism 36.

[0072] Optionally, the navigation path planning module 33 is used to plan a path based on the point information set, obtain the target movement path, and send the target movement path to the movement control module through the local control unit 35.

[0073] In an optional embodiment, the navigation path planning module 33 can autonomously calculate the safest optimal walking route within the factory area through the point information set issued by the task control center 2. Then, after the planning is completed, the robot can issue a walking command through the internal bus of the local control unit 35, thereby realizing the robot's local autonomous scheduling and reducing communication dependence.

[0074] For example, firstly, obtain all navigation-related parameters of the point information set, which may include the global coordinates of the target pump, the first target sampling station (recommended station), the backup sampling station, prohibited / restricted / accessible areas in the factory area, static obstacles, the robot's real-time battery level, real-time communication quality, and the estimated amount of sampling posture adjustment.

[0075] Secondly, all constrained areas can be marked on the built-in factory area navigation grid map to distinguish between high-risk restricted areas, ordinary restricted passages, and barrier-free routes. Furthermore, candidate sampling points that meet the sampling distance and sampling orientation constraints of the point database are screened, with the first target sampling station being prioritized and backup stations serving as alternative endpoints.

[0076] Then, the path cost for each candidate route is calculated. The cost dimension may include: path length, penalty for crossing dangerous areas, penalty for distance from obstacles, number of turns, communication signal attenuation, and attitude adjustment workload after arrival.

[0077] Furthermore, the costs of all candidate paths are compared, and the route with the lowest cost that satisfies explosion-proof safety, access constraints, and power constraints is selected as the target movement path. Additionally, if the first target sampling station is obstructed, inaccessible, or has excessive cost, a backup station can be automatically switched for replanning.

[0078] Finally, the target movement path is encapsulated into a local walking command package and sent to the local control unit 35 via the robot's internal local communication bus. Furthermore, the local control unit 35 caches the path information and sends it to the movement control module 34 as needed.

[0079] Optionally, the motion control module 34 is used to control the explosion-proof inspection robot 3 to move based on the target movement path, and when the explosion-proof inspection robot 3 moves to the first target sampling station, it sends a movement completion command to the local control unit 35, so that the local control unit 35 sends a visual acquisition command to the visual recognition module 31 based on the movement completion command.

[0080] In an optional embodiment, after the motion control module 34 receives the target movement path issued by the local control unit 35, it reads the robot's positioning coordinates, surrounding obstacle radar data, and explosion-proof safety sensor data in real time.

[0081] Furthermore, the mobile control module 34 can control the explosion-proof inspection robot 3 to travel at a constant speed along the planned path in segments based on the information read in real time, and avoid static obstacles, temporary personnel passage areas, and high-temperature corrosion high-risk areas in real time throughout the process.

[0082] Furthermore, the mobile control module 34 can continuously compare the real-time coordinates of the explosion-proof inspection robot 3 with the coordinates of the first target sampling station and calculate the coordinate deviation. Further, when both the coordinate deviation and the vehicle's parking position fall within the preset allowable range of the location database, it is determined that the robot has successfully reached the first target sampling station.

[0083] Furthermore, when the explosion-proof inspection robot 3 moves to the first target sampling station, the movement control module 34 sends a movement completion command to the local control unit 35. Then, under the control of the movement completion command, the local control unit 35 triggers visual acquisition and sends a visual acquisition command to the visual recognition module 31.

[0084] Optionally, the visual recognition module 31 is used to acquire the corresponding target area image based on the visual acquisition command, recognize the target area image to obtain the visual recognition result, and send the visual recognition result to the local control unit 35.

[0085] In one optional embodiment, when the visual recognition module 31 receives a visual acquisition command, it activates the built-in explosion-proof camera to acquire an image of the entire target pump area under the control of the visual acquisition command. The acquired image must cover the pump body, nameplate, QR code, pipeline interfaces, and preset visual marker areas.

[0086] Furthermore, matching algorithms can be adaptively selected based on the target visual features recorded in the location database. Specifically, if the acquired image contains a QR code / barcode, a decoding and recognition algorithm can be used; if the acquired image contains an equipment nameplate, image localization and OCR text extraction can be used; if the acquired image only contains the outline of a pump / pipeline, algorithms such as target detection, contour matching, key point matching, template matching, and semantic segmentation can be used.

[0087] Furthermore, all features extracted from image recognition can be compared item by item with the pre-stored visual profiles of target pumps in the location database. These features can include equipment number, QR code content, nameplate text, pump body outline, relative pipeline position, recommended sampling angle, etc.

[0088] Furthermore, the overall confidence level of image matching can be calculated based on the comparison results, and a visual recognition result can be generated, including the matching confidence level value, the identified device number, the sampling viewpoint deviation, and a mark indicating whether the match was successful. Finally, the obtained visual recognition result is sent to the local control unit 35.

[0089] Optionally, the local control unit 35 is used to determine whether the detection object contained in the target area image is the target key pump based on the visual recognition results and the key pump acoustic inspection point library 21, and when the detection object is the target key pump, it controls the explosion-proof inspection robot 3 to be in a low-noise parking sampling state based on the preset parking conditions, and sends an acoustic acquisition command to the acoustic acquisition module 32.

[0090] In an optional embodiment, after receiving the visual recognition result, the local control unit 35 retrieves the target pump location library visual reference data from the cached key pump acoustic inspection point library 21, and reads the matching confidence, equipment identification, and sampling angle from the visual recognition result.

[0091] Furthermore, the local control unit 35 compares the identification matching degree with the system's preset identity verification threshold. If the matching degree is greater than or equal to the preset threshold, the current detection object can be determined to be the target key pump; if the matching degree is less than the preset threshold, the current detection object can be determined not to be the target key pump.

[0092] Furthermore, if the current detection target is a key pump, i.e., the identity verification is successful, the local control unit 35 can also initiate a low-noise parking sampling pre-verification. The verification standard is shown in the following equation (1):

[0093] In the formula: Indicates the first Position, number The valid gate for parking in the sampling window requires all indicator functions to be true simultaneously to satisfy the parking condition. This indicates a conditional indicator function; the value is 1 if the condition inside the parentheses is true, and 0 if it is false. Indicates the speed of the robot chassis; This indicates that the chassis is completely stationary; This indicates the sampling direction deviation, which is the angular deviation between the actual orientation of the acoustic acquisition module and the standard sampling direction. This indicates the maximum allowable deviation threshold; Indicates the current actual sampling distance; This indicates the recommended sampling distance, which is the standard sampling distance recorded in the point database; Indicates the allowable distance error; This represents the noise energy of the robot itself; Indicates the upper limit threshold of the body noise; Indicates the confidence level of visual recognition; This represents the minimum confidence threshold for identity verification.

[0094] Furthermore, when the gate quantity When all conditions are met simultaneously, the robot enters a low-noise parking sampling state, which means that the chassis is locked, non-essential mechanisms such as the gimbal / robotic arm / fan reduce power or stop, and the vehicle's attitude is locked.

[0095] Furthermore, once the parking status is confirmed, the local control unit 35 sends an acoustic acquisition command to the acoustic acquisition module 32 to initiate the synchronous acquisition of multi-channel acoustic data.

[0096] Furthermore, when the gate quantity That is, if any condition is not met, no acquisition command will be issued. The position can be corrected by controlling the sampling posture adjustment mechanism 36 and then re-verified.

[0097] Optionally, the acoustic acquisition module 32 is used to acquire background sound field signals, robot body noise templates, acoustic signals of target key pumps and auxiliary state datasets based on acoustic acquisition commands, and send the background sound field signals, robot body noise templates, acoustic signals and auxiliary state datasets to the local control unit 35.

[0098] In an alternative embodiment, the acoustic acquisition module 32 may be an acoustic sensor employing a directional microphone, microphone array, or other suitable for explosion-proof environments.

[0099] In an optional embodiment, upon receiving an acoustic acquisition command, the acoustic acquisition module 32, under the control of the acoustic acquisition command, can synchronously and time-divisionally acquire three types of raw acoustic signals and auxiliary sensing data using an explosion-proof directional microphone / microphone array: ① Robot body noise template: The acoustic sensor is deflected to a position away from the target pump in no-load posture to collect the inherent noise generated by the chassis, motor, fan, and structural resonance; ② Background sound field signal: Collect ambient noise from adjacent pumps, fans, duct airflow, and exhaust equipment in the space outside the target; ③ Acoustic signals of key pumps and machines: The sensor is aligned with the faulty parts such as bearing housings / couplings / seals marked in the location database, and the sound and vibration of the pumps and machines are collected in a directional manner; ④ Auxiliary status dataset: Synchronously captured images from the equipment, infrared temperature, ambient gas concentration, and real-time process parameters (rotation speed). ,flow ,pressure medium temperature (valve position, load).

[0100] For example, setting Indicates the target key pump; Indicates the robot's sampling station location; Indicates the sampling window; Represents discrete sampling points; Indicates the frequency point. Then the robot at the [frequency point]. Raw acoustic sequences collected from each station It can be represented as the superposition of the target pump sound source, background sound field, robot body noise and random disturbance, as shown in the following relationship (2):

[0101] In the formula: This represents the sound source signal of a pure target pump; Indicates the time delay of sound wave propagation; Indicates the background sound field coupling term; This indicates the background sound field signals collected from adjacent equipment, pipeline reflections, airflow noise, etc. This represents the noise coupling term of the robot body; This represents the body noise template formed by the resonance of the robot chassis, motor, fan, and structure in this round of data collection; Indicates random perturbation; , , These represent the coupling weights of the target sound source, background sound field, and body noise at the current station location, respectively.

[0102] Furthermore, The coupling weight of the target pump sound source is determined by the sampling distance, orientation deviation, and pose, as shown in the following equation (3):

[0103] In the formula: This indicates the deviation angle between the direction the acoustic acquisition module points to and the normal or recommended sampling direction of the target area; Indicates the sampling distance; This indicates the near-field distance correction amount; Indicates the distance attenuation coefficient; This indicates the actual sampling pose of the robot or sampling posture adjustment mechanism. This represents the recommended sampling pose recorded in the point database; This represents the attitude deviation weight matrix.

[0104] If the explosion-proof inspection robot 3 is equipped with a robotic arm, then This can include the position and orientation of the robotic arm's end effector; if the explosion-proof inspection robot 3 is not equipped with a robotic arm, then... The attitude can be determined by the posture of the vehicle body, gimbal, lifting mechanism, or telescopic mechanism.

[0105] Furthermore, the collected background sound field signal, robot body noise template, acoustic signal and auxiliary state dataset are sent to the local control unit 35.

[0106] Optionally, the local control unit 35 is also used to encapsulate the background sound field signal, the robot body noise template, the acoustic signal and the auxiliary state dataset to obtain a complete set of sampling datasets, and to perform local basic credibility judgment on the complete set of sampling datasets, and when the local basic credibility meets the first preset threshold requirement, to send the complete set of sampling datasets to the task control center 2.

[0107] In an optional embodiment, after receiving the background sound field signal, robot body noise template, acoustic signal and auxiliary status dataset, the local control unit 35 can perform unified time alignment, format standardization and add metadata such as current sampling station, robot number and task number to the four types of raw data, and then encapsulate them to generate a complete set of sampling datasets.

[0108] Furthermore, the local control unit 35 can also utilize only shallow indicators that can be read locally and in real time by the robot, using gating quantities. The system uses built-in conditions such as chassis static state, sampling distance deviation, orientation angle deviation, body noise energy, and visual recognition confidence level to make local basic credibility judgments.

[0109] Furthermore, the obtained local basic trustworthiness metric is quantified into a simplified score and compared with a system-preset first preset threshold. If the local basic trustworthiness is greater than or equal to the first preset threshold, the sampling basic conditions are deemed acceptable, and the entire sampling dataset is then uploaded to the task control center 2 via encrypted communication.

[0110] Furthermore, if the local basic credibility is less than the first preset threshold, the sampling quality can be determined to be unqualified, and data will not be transmitted to the task control center 2.

[0111] Optionally, the local control unit 35 is also used to send a first adjustment command to the sampling attitude adjustment mechanism 36 when the detected object is not the target critical pump.

[0112] In an optional embodiment, when the object being detected is not the target critical pump, the local control unit 35 can also send a first adjustment command to the sampling posture adjustment mechanism 36. This first adjustment command may include corrections to the target pose, angle adjustment range, distance adjustment amount, and gimbal / robotic arm motion constraints.

[0113] The process of determining whether the detection object is the target key pump can be referred to in the previous text for the relevant functional description of the ground control unit 35, and will not be repeated here.

[0114] Optionally, the sampling posture adjustment mechanism 36 is used to adjust the position and orientation of the explosion-proof inspection robot 3 based on the first adjustment command, and send the first adjustment completion command to the local control unit 35 according to the adjusted position and orientation, so that the local control unit 35 sends a new visual acquisition command to the visual recognition module 31 based on the first adjustment completion command.

[0115] In an alternative embodiment, a robotic arm may be provided, but the robotic arm is not the only necessary condition for the system to operate.

[0116] Furthermore, the robotic arm serves as an optional implementation of the sampling posture adjustment mechanism, used to further adjust the position of the acoustic acquisition module or auxiliary sensor relative to the target detection area after the robot has reached the target sampling station.

[0117] Furthermore, in embodiments without a robotic arm, the robot can adjust the orientation, height, and distance of the acoustic acquisition module through vehicle steering, gimbal rotation, lifting mechanism, or telescopic mechanism to achieve non-contact near-field acoustic acquisition.

[0118] Furthermore, in the embodiment where a robotic arm is configured, the robotic arm can move the acoustic acquisition module 32 to a position close to the pump body, bearing housing, coupling area, mechanical seal area, inlet pipe section, outlet pipe section, or valve group area according to the target detection location determined by the vision recognition module 31, so as to obtain acoustic signals closer to the target location.

[0119] Furthermore, the robotic arm is primarily used for proximity sampling, orientation adjustment, or auxiliary detection, and is not required to perform disassembly, repair, sealing, or other high-risk operations in the main process. If a contact-type auxiliary sensor is added to the subsequent system, the robotic arm can also briefly approach or contact the target detection area for supplementary sampling, provided that remote authorization and safety conditions are met.

[0120] In an optional embodiment, upon receiving a first adjustment command, the sampling posture adjustment mechanism 36 adjusts the position and orientation of the explosion-proof inspection robot 3 under the control of the first adjustment command. The adjustment process can be performed based on different hardware types. 1. No robotic arm configuration: Controls vehicle steering and uses a lifting / extending gimbal to change the orientation and height of the acoustic acquisition module; 2. With robotic arm configuration: Synchronously adjust the position of the end effector of the robotic arm to correct the relative observation angle between the vision camera and the pump body.

[0121] Furthermore, during the adjustment process, the pose sensor data can be read in real time, and the difference between the actual pose and the target pose can be continuously verified until the deviation falls within the allowable range of the point library. Then, the pose correction is completed and a first adjustment completion command containing the new vehicle orientation, gimbal angle, robotic arm pose, and sampling distance is generated.

[0122] Furthermore, the first adjustment completion instruction is sent to the local control unit 35, and then, under the control of the first adjustment completion instruction, the local control unit 35 sends a new visual acquisition instruction to the visual recognition module 31 and re-captures the target area image to perform identity matching.

[0123] Furthermore, the system iteratively executes logic for identifying mismatches, adjusting posture, and retrying until the matching accuracy meets the standard or the maximum number of retries is reached. Additionally, if the retries exceed the limit, an exception log can be uploaded to Task Control Center 2 to request manual review.

[0124] Optionally, the local control unit 35 is also used to send a second adjustment command to the sampling posture adjustment mechanism 36 when the local basic confidence does not meet the first preset threshold requirement.

[0125] In an optional embodiment, the above relation (1) can be used to determine the local basic credibility. Furthermore, if the deviation is only in the sampling angle and sampling distance, and the chassis is stationary, the body noise, and the visual recognition are all qualified, then it is determined that the sampling quality can be repaired by fine-tuning the posture in place, and a second adjustment command is generated.

[0126] The second adjustment command can carry information such as angle correction, sampling distance fine-tuning parameters, and gimbal lifting amplitude, and is used to optimize only the pointing of the acoustic acquisition module without changing the robot's docking point.

[0127] Furthermore, the second adjustment command is sent to the sampling posture adjustment mechanism 36.

[0128] Optionally, the sampling posture adjustment mechanism 36 is also used to adjust the position and orientation of the explosion-proof inspection robot 3 based on the second adjustment command, and send the second adjustment completion command to the local control unit 35 according to the adjusted position and orientation, so that the local control unit 35 sends a new acoustic acquisition command to the acoustic acquisition module 32 based on the second adjustment completion command.

[0129] In an optional embodiment, upon receiving a second adjustment command, the sampling attitude adjustment mechanism 36 adjusts the position and orientation of the explosion-proof inspection robot 3 under the control of the second adjustment command, and recalculates the target pump sound source coupling weight using the above-mentioned relationship (3) after correction. The adjustment process can be referred to in the previous description of the relevant functions of the sampling attitude adjustment mechanism 36, and will not be repeated here.

[0130] Furthermore, once the attitude adjustment is complete, the gimbal, robotic arm, and vehicle body are locked in place to maintain stability, and a second adjustment completion command is generated and sent back to the local control unit 35. Subsequently, under the control of this second adjustment completion command, the local control unit 35 issues a new acoustic acquisition command to the acoustic acquisition module 32, and uses the acoustic acquisition module 32 to resynchronize and acquire the background sound field, the body noise template, the target pump acoustic signal, and the auxiliary working condition dataset.

[0131] Furthermore, after resampling, the local basic credibility judgment is performed again. If it meets the standard, the entire set of sampling dataset is packaged and uploaded to the task control center 2; if it still does not meet the standard, the logic of switching to the backup sampling station is triggered.

[0132] Optionally, the local control unit 35 is also used to send a movement command to the movement control module 34 when the local basic credibility does not meet the first preset threshold requirement, so that the movement control module 34 controls the explosion-proof inspection robot 3 to move to the second target sampling station or the backup sampling station based on the movement command.

[0133] In one optional embodiment, if the local basic reliability is still lower than the first preset threshold after two in-situ attitude resampling operations by the local control unit 35, it can be determined that the current recommended station cannot obtain effective acoustic and vibration data.

[0134] Furthermore, the local control unit 35 can reread the information such as the coordinates of the backup sampling station, the standard sampling direction of the backup station, the sampling distance, and the distribution of surrounding interference sound sources in the key pump acoustic inspection point library 21, and generate a movement command that includes the coordinates of the target backup station, the safe movement path, and the standard sampling posture after arrival.

[0135] Furthermore, the generated movement command is sent to the movement control module 34, which then controls the explosion-proof inspection robot 3 to move to the second target sampling station or a backup sampling station under the control of the movement command. The specific control process can be found in the functional description of the movement control module 34 above, and will not be repeated here.

[0136] Furthermore, after the robot travels to the backup sampling station (the second target sampling station), it repeatedly performs the identification and verification of the device identity, parking and gate control verification, acoustic multi-type data acquisition, local credibility judgment, and finally obtains the multi-station re-sampling results.

[0137] Optionally, the mission control center 2 is also used to generate anomaly review tasks based on fixed sensor alarms, DCS operating condition anomalies, manual remote commands, or historical trend anomalies, and to determine the target key pumps based on the anomaly review tasks.

[0138] In an optional embodiment, when any one of the following channels—fixed online monitoring sensors in the plant area, process DCS system, manual operation by maintenance personnel, or historical diagnostic time-series data—raises clues about equipment anomalies, the task control center 2 can proactively generate a targeted review task and prioritize dispatching the explosion-proof inspection robot 3 to the corresponding pump to conduct a special verification of acoustic vibration.

[0139] Furthermore, through the above operations, the shortcomings of fixed monitoring, which only uses single-point sensing and cannot locate the root cause of mechanical failures, are made up for, and it is possible to quickly distinguish between real equipment failures and normal process fluctuations and sensor false alarms.

[0140] Optionally, the mission control center 2 is also used to acquire historical acoustic and vibration characteristics that continuously deviate from the baseline, and to generate fault trend tracking and diagnostic tasks based on the operating conditions.

[0141] In an optional embodiment, by continuously comparing the historical acoustic and vibration characteristics stored in multiple rounds of inspections with the corresponding operating condition baseline, and identifying long-term stable deviation trends, the task control center 2 can independently generate a low-frequency long-cycle trend tracking task, which can continuously observe the equipment degradation rate, predict potential major leakage and explosion risks in advance, and make up for the shortcomings of a single inspection that only provides instantaneous status assessment.

[0142] Optionally, the mission control center 2 is also used to perform data processing, sound field compensation, adaptive noise suppression, multi-station sound source attribution discrimination, diagnostic calculation and operating condition baseline matching on the complete set of sampled datasets in sequence, and obtain the target fault diagnosis results of the target key pump.

[0143] In an alternative embodiment, the task control center also includes a data analysis engine 22 and a diagnostic results database 23.

[0144] Optionally, the data analysis engine 22 is used to perform a comprehensive confidence assessment of the entire set of sampled datasets in all dimensions. When the comprehensive confidence assessment in all dimensions meets the second preset threshold requirement, the engine sequentially performs data processing, sound field compensation, adaptive noise suppression, multi-station sound source attribution discrimination, diagnostic calculation and working condition baseline matching on the entire set of sampled datasets, and obtains the target fault diagnosis results of the target key pump.

[0145] In an optional embodiment, the data analysis engine 22 can extract seven basic input indicators from the complete set of sampling datasets, the key pump acoustic inspection point library 21, and the multi-station re-sampling results, and calculate the comprehensive credibility across all dimensions using the following relation (4):

[0146] In the formula: Indicates the first The overall reliability of each sampling window is in all dimensions, with a value range of (0, 1). The higher the value, the better the quality of the sampled data. This represents the Sigmoid normalization function, which is used to map the linear weighted result to the interval 0~1 to avoid unbounded scores. This represents a constant bias term, which is a fixed constant preset by the system. This represents the confidence weight of visual recognition, with positive scores indicating higher confidence levels; the higher the matching degree, the higher the confidence. This indicates the weight of the parking gate control quantity, with positive bonuses. All parking conditions must be met to score a point. ; The target sound signal-to-noise ratio weight is positively scored, with a higher score for a higher proportion of effective sound signal. This represents the probability weight of the target pump's sound source attribution, with positive scoring; the more certain the anomaly is to belong to the target pump, the higher the score. This indicates a negative weighting of the background sound field energy; the greater the background noise, the more the credibility is reduced. This indicates a negative weight for the robot's own noise; the greater the robot's self-noise, the more the credibility is reduced. This indicates a negative weighting for operating condition fluctuations; the greater the change in operating conditions before and after sampling, the greater the reduction in reliability. This indicates the confidence level of the visual recognition match in the current window, taken from the visual recognition results uploaded by the robot; The parking sampling threshold is represented by the above formula (1) and is calculated. This represents the signal-to-noise ratio of the target pump's acoustic and vibration signals after acoustic field compensation and adaptive noise suppression. The output of the multi-station sound source attribution judgment is the probability score of the abnormality belonging to the target pump, which is calculated by the above relationship (5); This represents the total energy of the background sound field signal in the current sampling window; This represents the total energy of the robot's body noise template in the current sampling window; This represents the real-time operating condition vector for the current sampling window.

[0147] Furthermore, the calculated comprehensive credibility across all dimensions is then... Compare with the second preset threshold; if the overall credibility across all dimensions is... If the sampled data is greater than or equal to the second preset threshold, it can be determined that the quality of the sampled data is qualified in all dimensions, and then the data processing, sound field compensation, adaptive noise suppression, multi-station sound source attribution determination, diagnostic calculation, and working condition baseline matching process are executed in sequence.

[0148] Furthermore, if the overall credibility is considered... If the data is less than the second preset threshold, it can be determined that the overall data interference is too large and the reliability is insufficient, and the fault diagnosis operation will not be performed. Furthermore, the data analysis engine 22 can generate sampling verification instructions and send them to the task control center 2. The task control center 2 then schedules the explosion-proof inspection robot 3 to perform in-situ attitude re-sampling or switch to a backup station for re-sampling. At the same time, low-confidence data is marked and stored in the diagnostic result database 23 for future reference.

[0149] Furthermore, the entire set of sampled datasets undergoes standardized preprocessing to eliminate computational interference caused by the original acquisition format, temporal misalignment, and abnormal segments. Additionally, sound field compensation eliminates the attenuation effects of sampling distance, sampling angle, and pose deviations on the amplitude of the pump's acoustic vibration signal, ensuring lateral comparability of data acquired from different stations and postures. Finally, adaptive noise suppression layering effectively separates background environmental noise and inherent robot noise, enabling the extraction of a pure target pump acoustic vibration response signal.

[0150] Furthermore, by distinguishing whether the source of the abnormal spectrum signal is the target pump or a neighboring pump, fan, valve, or other interfering equipment, false alarms caused by crosstalk from multiple devices in the plant can be resolved. Moreover, by extracting fault-sensitive multidimensional acoustic and vibration features, the degree of abnormality of equipment deviating from its healthy state can be quantified, and standardized fault quantification indicators can be output.

[0151] Finally, by matching the operating condition baseline, the influence of process condition fluctuations such as speed, flow rate, pressure, and temperature on the acoustic and vibration signals can be eliminated, avoiding misjudging normal operating condition changes as equipment failures, and ultimately outputting a complete set of diagnostic conclusions, namely the target fault diagnosis results, including equipment health level, fault type, fault location, and risk level.

[0152] Furthermore, the obtained target fault diagnosis results are stored in the diagnosis result database 23.

[0153] This embodiment provides a method for diagnosing critical pump failures, which can be used in the task control center 2 of the aforementioned critical pump failure diagnosis system 1. Figure 2 This is a flowchart of a key pump fault diagnosis method according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S201: Receive the complete set of sampling datasets of the target key pump sent by the explosion-proof inspection robot in the key pump fault diagnosis system.

[0154] For the specific process, please refer to the functional description and specific process description of the task control center 2 and the explosion-proof inspection robot 3 in the above-mentioned key pump fault diagnosis system 1, which will not be repeated here.

[0155] Step S202 involves processing the entire set of sampled datasets, performing sound field compensation, adaptive noise suppression, and multi-station sound source attribution to obtain the compensated acoustic vibration response feature set.

[0156] In one optional embodiment, the acoustic and vibration response feature set represents a clean, unified, and standardized set of target pump frequency domain / time domain acoustic and vibration indicators obtained after data cleaning, adaptive noise suppression, multi-station sound source attribution discrimination, and posture-consistent sound field compensation, stripping away robot self-noise, factory environment noise, and crosstalk interference from adjacent equipment.

[0157] In one optional embodiment, the entire sampled dataset is first parsed, formatted uniformly, time-synchronized, and abruptly removed. Then, by performing a short-time frequency domain transformation on the original signal, background sound field, and robot body noise, the following parameters are obtained: , , Three types of frequency domain data.

[0158] Furthermore, substituting the adaptive suppression factor formula into the frequency subtraction of background and robot body noise, the following relationships (5) and (6) are shown:

[0159] In the formula: This represents the frequency domain acoustic and vibration response signal of the clean target pump after adaptive noise suppression is completed; Indicates the adaptive noise suppression factor; This represents the frequency domain amplitude of the original mixed acoustic signal collected by the explosion-proof inspection robot 3 after short-time Fourier transform; This represents the background sound field frequency compensation coefficient, which varies with frequency and is used to quantitatively subtract ambient noise components at the corresponding frequency points. This represents the frequency domain amplitude of the acquired background sound field signal; This represents the robot's own noise frequency compensation coefficient, which varies with frequency and quantitatively deducts the noise from the robot's own motor, fan, and structural resonance. This represents the frequency domain amplitude corresponding to the robot's body noise template; This represents a fixed weighting coefficient for background sound field interference, used to characterize the intensity of environmental noise interference on the signal; This represents the fixed weighting coefficient for robot body noise, used to characterize the intensity of robot self-noise interference; This represents a minimal stability constant, used to prevent computational overflow and numerical distortion caused by the denominator approaching 0.

[0160] Furthermore, The value range is 0~1. Furthermore, when the target pump signal energy is much greater than the background and robot noise, It fully preserves the target's acoustic and vibration signals; when the background / robot noise energy accounts for a very high proportion, Approaching 0 significantly reduces the signal weight of interference frequencies, suppressing erroneous features caused by noise.

[0161] Furthermore, by substituting multiple sets of station compensation signals into the residual and probability formulas, it is possible to distinguish whether the anomaly originates from the target pump and filter out external interference sound source data, as shown in the following relationships (7) and (8):

[0162] In the formula: Indicates candidate sound sources The smaller the residual, the higher the degree of matching between the sound source and the sound energy attenuation pattern at multiple stations. This represents the set of all sampling stations; Indicates the first Position weight; This indicates the energy of the anomalous frequency band after compensation. Indicates candidate sound sources To the Standing distance; Indicates the near-field correction amount; This indicates that the abnormal sound source belongs to the sound source category. Probability score; This represents the set of all surrounding candidate sound sources.

[0163] Furthermore, the differences in signal amplitude at different stations are corrected based on sampling distance, angle, and pose to eliminate energy deviation caused by sampling posture. Then, by summarizing all valid, interference-free, and standardized frequency and time domain indicators, an acoustic vibration response feature set is generated.

[0164] For example, the basic process of sound field compensation processing may include: 1. Estimate the current ambient noise level based on the background sound field signal to reduce the impact of adjacent equipment, pipeline reflections, airflow noise, and ambient noise on the target signal; 2. Based on the robot body noise template, subtract or suppress interference introduced by the robot chassis, motor, fan, power module and structural resonance; 3. Standardize the data from different inspection cycles based on sampling posture, sampling distance, and sampling direction; 4. Based on the sampling results from multiple stations, determine whether the abnormal sound source comes from the target pump, and mark or trigger re-sampling for data with insufficient credibility.

[0165] Furthermore, the compensated acoustic and vibration response characteristics can include sound pressure level changes, root mean square value, peak value, impact characteristics, frequency band energy, dominant frequency component, octave relationship, envelope characteristics, spectral peak changes, spectral centroid, proportion of abnormal frequency band energy, and time trend changes. Moreover, these characteristics are primarily used to describe the acoustic state of pump operation, and no single formula is used as the sole basis for system diagnosis.

[0166] Step S203: Extract indicators from the acoustic vibration response feature set and construct an acoustic vibration feature vector set.

[0167] In one optional embodiment, the acoustic vibration feature vector set represents a standardized vector obtained by extracting multi-dimensional quantized values ​​from the acoustic vibration response feature set according to a preset fault-sensitive frequency band and time-domain impact index, and encapsulating all fault characterization indicators from a single round of sampling. Each round of sampling corresponds to a set of independent feature vectors, and multiple rounds of sampling are aggregated to form a vector set used for baseline comparison, anomaly quantification, and fault classification.

[0168] In one alternative embodiment, all fault-sensitive quantitative indicators are extracted from the compensated pure acoustic-vibration response and encapsulated into a single-round feature vector, which is then aggregated through multiple rounds of sampling vectors to form a vector set.

[0169] For example, after compensation, according to the fault-sensitive frequency band set Extract acoustic vibration energy characteristics. Further, The frequency can be set according to the rotational frequency, frequency multiplication factor, bearing characteristic frequency band, coupling abnormal frequency band, cavitation high-frequency impact frequency band, or empirical frequency band, as shown in the following relationship (9):

[0170] In the formula: Indicates the first The normalized energy percentage of the fault-type frequency band; the higher the value, the more obvious the corresponding fault characteristics. The fault frequency band number corresponds to specific frequency bands for various faults such as bearings, rotor imbalance, coupling misalignment, cavitation, and foundation loosening. Indicates the first Within each fault frequency band, frequency The corresponding frequency band weights are used to amplify the contribution of the core fault frequency and weaken irrelevant frequencies; This represents the complete total frequency range of all data involved in the analysis; This represents the minimum stability constant, used to prevent calculation overflow when the total energy of the denominator approaches 0.

[0171] Furthermore, the simultaneous extraction of global integrated time-domain / frequency-domain features may include the spectral centroid. Sideband aggregation High-frequency impact density Envelope modulation intensity This allows all the indicators from a single round of sampling to be combined into a single acoustic vibration feature vector. The following relation (10) is shown:

[0172] In the formula: express Normalized energy of different fault frequency bands ; It represents the centroidal characteristics of the spectrum and is used to characterize the overall offset position of acoustic and vibrational energy; It indicates the degree of edge band aggregation and is used to identify periodic faults such as shaft misalignment and rotor imbalance. It represents the high-frequency impact density characteristics and is used to identify impact-related damage such as pitting and cavitation in bearings; It represents the envelope modulation intensity characteristics and is used to characterize periodic modulation noise caused by poor bearing lubrication and seal friction.

[0173] Furthermore, the current vector is stored in a time-series cache and combined with historical sampling vectors from multiple rounds to form an acoustic-vibration feature vector set.

[0174] Step S204: Based on the key pump acoustic inspection point library, construct a real-time operating condition vector, and based on the real-time operating condition vector and acoustic vibration feature vector, construct a comprehensive acoustic vibration anomaly index.

[0175] In one optional embodiment, the comprehensive acoustic and vibration anomaly index represents a normalized comprehensive quantitative index of the overall reliability of fused sampling, the degree of deviation from normalized operating conditions, the fault energy of each frequency band, the deterioration upward trend, and the historical health similarity. The larger the value, the higher the degree of mechanical deterioration of the pump.

[0176] In one optional embodiment, considering that the normal acoustic state of the pump varies under different speeds, loads, flow rates, pressures, temperatures, and valve positions, a real-time operating condition vector is constructed by combining a database of key pump acoustic inspection points. The following relation (11) is shown:

[0177] In the formula: This indicates the real-time operating load of the motor; This indicates the real-time opening degree of the matching inlet and outlet valves.

[0178] Furthermore, by using operating conditions to weight and match historical healthy baselines, and calculating the baseline mean, covariance, operating condition normalization deviation, and deterioration trend layer by layer, a comprehensive acoustic and vibration anomaly index can be finally obtained through weighted fusion.

[0179] Step S205: Determine the foundation condition level of the target key pumps based on the comprehensive acoustic and vibration anomaly index.

[0180] In one optional embodiment, the basic status level represents a four-level equipment health status based on the comprehensive acoustic and vibration anomaly index range, which may include normal, attention, abnormal, and severe abnormal.

[0181] In one optional embodiment, the system presets four comprehensive acoustic-vibration anomaly index threshold ranges, each corresponding to a basic state level: 1. Range 1 (Comprehensive acoustic and vibration anomaly index is extremely low): Normal, no mechanical deterioration of the equipment, only routine periodic inspection; 2. Interval 2 (low value): Pay attention, slight feature shift, will be automatically reviewed in the next round of inspection; 3. Range 3 (Medium value): Abnormal, with clear mechanical damage, remote alarm pushed, and key maintenance follow-up; 4. Interval 4 (high value): serious abnormality, high risk of failure, automatic generation of maintenance work order, and linkage with process system verification.

[0182] Furthermore, the calculated comprehensive acoustic and vibration anomaly index is compared with the above four thresholds in sequence, and a unique basic state level is output after matching the corresponding interval.

[0183] Step S206: Based on the comprehensive acoustic and vibration anomaly index, acoustic and vibration feature vector set, real-time operating condition vector, and basic state level, fault diagnosis is performed using preset fault type acoustic and vibration judgment rules to obtain the target fault diagnosis result of the target key pump.

[0184] In one optional embodiment, the preset fault type acoustic vibration determination rules are shown in Table 2 below: Table 2. Rules for Judging Acoustic Vibration by Preset Fault Types

[0185] Furthermore, the energy of each fault frequency band within the acoustic vibration characteristic vector is used as a basis for further analysis. , , , , Using real-time operating condition vectors as the core matching criterion Eliminate interference from process fluctuations.

[0186] Furthermore, the system integrates the comprehensive acoustic and vibration anomaly index and the basic condition level to differentiate the severity of the fault, and outputs a complete target fault diagnosis result after matching is completed. This target fault diagnosis result can further include: the equipment's basic condition level; suspected fault type and high-incidence fault location; comprehensive acoustic and vibration anomaly index and normalized deviation of operating conditions; anomaly upward trend and sampling reliability; and corresponding maintenance and repair recommendations (automatic verification / remote viewing / emergency maintenance).

[0187] Furthermore, all diagnostic results can be written into the diagnostic results database 23 for archiving, and hierarchical alarms, inspection reports, and maintenance work orders can be generated simultaneously to complete the diagnostic closed loop.

[0188] The key pump fault diagnosis system provided in this embodiment solves the industry pain points of high safety risks of manual inspection in hazardous chemical plants, high deployment costs of fixed sensors, and the inability of ordinary robots to identify early acoustic and vibration faults of pumps through a collaborative mechanism of robot-based on-site quality control and multi-layer noise reduction and source tracing diagnosis in the background. It achieves stable output of highly reliable fault judgment results without deploying a large number of contact sensors.

[0189] In some optional implementations, step S204 above, which constructs a comprehensive acoustic-vibration anomaly index based on real-time operating condition vectors and acoustic-vibration feature vectors, includes: Step a1: Determine the mean and covariance matrix of the normal acoustic and vibration baseline that matches the operating conditions of the target key pump based on the real-time operating condition vector.

[0190] Step a2: Based on the acoustic vibration feature vector set, the mean and covariance matrix of the normal acoustic vibration baseline, determine the normalized abnormal deviation of the working condition.

[0191] Step a3: Construct a comprehensive acoustic and vibration anomaly index based on the normalized anomaly deviation of the working conditions.

[0192] In one optional embodiment, the mean of the normal acoustic vibration baseline represents the average value of the feature vector of historical healthy samples that closely matches the current real-time operating conditions, representing the standard fault-free acoustic vibration level of the pump under this operating condition; the covariance matrix represents the multidimensional covariance matrix of the historical health characteristics under the same operating conditions, which can characterize the reasonable fluctuation range of each acoustic vibration index under normal operation and is used to distinguish between normal fluctuations under operating conditions and real mechanical faults.

[0193] Furthermore, the normalized deviation of the operating condition represents the quantized Mahalanobis distance between the current acoustic vibration feature vector and the mean and fluctuation range of the healthy baseline under the same operating condition. This is used to eliminate signal differences caused by changes in process conditions such as speed, flow rate, pressure, and temperature. Furthermore, only feature offsets caused by mechanical damage are retained; the larger the offset, the more severe the fault.

[0194] In one alternative embodiment, all historical healthy samples in the database of key pump acoustic inspection points are first retrieved. Then, the matching weight between the current operating condition and each historical operating condition is calculated, as shown in the following equation (12):

[0195] In the formula: Indicates the first The historical baseline sample and the current number The working condition matching weight of the window working condition has a value range of (0,1]. Furthermore, the closer the current working condition is to the historical sample working condition, the closer the weight is to 1. The greater the difference in working conditions, the more the weight decays to 0. This represents the natural exponential function, used to implement Gaussian similarity weighting; Indicates the first Historical working condition vectors corresponding to each historical health baseline sample; This represents the vector of differences between the current operating conditions and historical sample operating conditions. The operating condition sensitive weight matrix is ​​a diagonal matrix, whose diagonal elements correspond to the sensitivity coefficients of speed, flow rate, pressure, temperature, load, and valve position, respectively. It is used to amplify the differences in process parameters that have a significant impact on acoustic and vibration signals and to weaken the fluctuations of irrelevant parameters. Furthermore, the weighted calculation of the mean and covariance matrix of the health baseline under the same working conditions is shown in the following equations (13) and (14):

[0196] In the formula: This represents the mean vector of normal acoustic and vibration baselines obtained after weighted matching under the current real-time operating conditions, and represents the standard acoustic and vibration characteristic center value when the pump is fault-free under this operating condition. This represents the acoustic and vibrational feature vector corresponding to the historical baseline sample; This represents a weighted sum of all historical feature vectors based on their similarity to the operating conditions. This represents the sum of the weights of all historical sample operating conditions, used to normalize the weighted mean. It represents the normal acoustic and vibration baseline covariance matrix under the current operating conditions, which can characterize the fluctuation amplitude of multi-dimensional acoustic and vibration features under healthy conditions and the correlation between features, and is used to distinguish between normal operating condition fluctuations and mechanical fault deviations. This represents the deviation vector between the features of a single historical sample and the mean of the baseline for matching working conditions. This represents the outer product of the bias vectors, used to generate a multidimensional feature covariance matrix; Furthermore, Mahalanobis distance is used to eliminate differences in operating conditions and to quantify the degree to which the current characteristics deviate from the healthy baseline, as shown in the following relationship (15):

[0197] In the formula: This indicates the abnormal deviation of the normalized operating conditions. The larger the value, the more significant the deviation of the current acoustic and vibration characteristics from the historical normal state under the same working conditions; This represents the current acoustic vibration eigenvector; This represents a regular stable term.

[0198] Furthermore, the abnormal upward trend index characterizing the rate of continuous equipment degradation is calculated as shown in the following equation (16):

[0199] In the formula: This indicates the upward trend of the abnormal deviation, i.e., the value of the upward trend of the deviation. This represents the exponential moving average.

[0200] Furthermore, a comprehensive acoustic-vibration anomaly index is generated through multi-dimensional weighted fusion. The following relation (17) is shown:

[0201] In the formula: , , , Both represent global weights; Indicates the fault band weight; Indicates historical health similarity.

[0202] In some optional implementations, the above method further includes: Step b1: Generate a graded closed-loop handling plan based on the target fault diagnosis results.

[0203] In one optional embodiment, by utilizing the comprehensive acoustic and vibration anomaly index, basic condition level, fault type, comprehensive sampling reliability, and deterioration upward trend results from the diagnostic output, differentiated handling processes are divided according to the level of risk. This can automatically generate standardized closed-loop handling solutions, connecting the entire chain of fault identification, alarm push, manual review, work order dispatch, process linkage, and trend tracking. It eliminates the manual judgment link in fault handling and can thus adapt to the safety management requirements of pumps and machinery in hazardous chemical plants.

[0204] For example, firstly, when the basic status level is severe anomaly, Higher than the high-risk threshold; or When the risk of a sustained and significant increase in pollution levels occurs, and the pump is a high-risk medium feed / return pump; and the fault type is a mechanical seal leak, bearing seizure, severe cavitation, or other faults that are prone to leakage and explosion, the corresponding high-risk handling plan is as follows: 1. Real-time hierarchical alarm: Simultaneously send three levels of emergency alarms to the plant process control center, equipment operation and maintenance manager, and safety administrator via sound and light, SMS, and platform pop-up window, and attach the fault pump number, unit area, fault type, anomaly index, and on-site sampled sound and vibration spectrum; 2. Automatically generate maintenance work orders: The work order is bound to the equipment location, fault location, historical baseline comparison curve, sampled full set of dataset attachments, and recommended maintenance items (bearing replacement, seal overhaul, rotor dynamic balance correction, etc.), and is automatically pushed to the operation and maintenance dispatch system; 3. Process system linkage: Push early warning signals to the DCS system and provide two optional linkage strategies: short-term load reduction operation and equipment isolation standby pump switching prompt, reminding process personnel to monitor medium pressure and flow fluctuations; 4. Forced encrypted tracking task: Automatically adds high-frequency fault trend tracking and diagnosis tasks, shortens the inspection cycle, and the robot completes multi-station verification sampling every 2 hours; 5. Mandatory event archiving: Synchronously record the time of fault occurrence, alarm push records, and work order issuance time, and store them permanently in the diagnostic results database for use in tracing the source of safety incidents.

[0205] Secondly, when the basic status level is abnormal, It is in the middle range and shows no trend of rapid deterioration. When the faults are mostly minor imbalances, slight pipe blockages, or minor localized cavitation, the corresponding medium-risk handling plan is as follows: 1. Platform-based alarm push: Alarm entries are generated only in the equipment operation and maintenance management backend, with accompanying sound and vibration characteristic comparison charts, and do not trigger emergency SMS pushes; 2. Generate a pending review reminder: Mark the equipment on the daily inspection priority list of maintenance personnel, requiring on-site manual listening and infrared temperature measurement review to be completed within 3 working days; 3. Schedule routine verification and inspection tasks: In the next round of scheduled inspection tasks, automatically upgrade the pump to high priority and force the robot to use multi-station sampling and verification; 4. Continuous Trend Monitoring: Maintain the original inspection cycle, but continuously record each round of data. , If the index rises for two consecutive rounds in time-series data, it will automatically be upgraded to a serious anomaly handling procedure.

[0206] Then, when the base status level is "Attention", A slight deviation from the healthy baseline, without a sustained upward trend, likely indicates minor fluctuations in operating conditions or early, very slight wear. The corresponding low-risk handling plan is as follows: 1. Internal data tagging only: The diagnostic results are tagged with relevant tags within the database, without proactive alarm pushes, so as not to disturb operation and maintenance personnel; 2. No additional special tasks: No new review or trend tracking tasks will be added; the existing fixed-cycle inspections will be used for continuous observation. 3. Baseline dynamic update judgment: If the index falls back to the normal range in subsequent rounds of sampling, the watch label will be automatically removed; if the index rises slowly, it will be automatically upgraded to the medium-risk anomaly handling process.

[0207] Finally, when the base status level is normal, When the condition is within the healthy range, and its characteristics highly overlap with the baseline operating conditions, the corresponding risk-free and normal handling procedure is as follows: 1. Archive-only storage: Store the complete set of sampling data and diagnostic indicators in the database without generating any alarms, work orders, or review tasks; 2. Participate in health baseline iteration: Weight and update the corresponding working conditions using the qualified health samples from this study. , Baseline parameters are continuously optimized to determine subsequent diagnostic criteria.

[0208] In one example, a critical pump and machinery acoustic and vibration condition monitoring and intelligent diagnostic system and method are provided. This solution is applied to production areas, storage and transportation areas, pipe corridors, and other locations in hazardous chemical enterprises where there is a need to transport flammable, explosive, toxic, harmful, or corrosive media. The monitored objects are mainly critical pumps and machinery that operate continuously or have high-frequency start-stop cycles, including media transfer pumps, circulating pumps, cooling pumps, reaction feed pumps, reflux pumps, compressor-supporting pump sets, and moving equipment related to important process materials.

[0209] Furthermore, this solution utilizes an explosion-proof inspection robot as a mobile platform, employs acoustic signal acquisition and acoustic vibration state analysis as the primary detection methods, and integrates visual recognition, path navigation, operating condition data, historical baselines, and a diagnostic platform to conduct non-contact or near-field condition monitoring of critical hazardous chemical pumps. Here, "acoustic vibration" refers to characterizing the pump's vibration state and abnormal mechanical trends by collecting acoustic signals such as sound, impact sound, friction sound, cavitation sound, and periodic mechanical sound generated during pump operation.

[0210] Furthermore, such as Figure 3 and Figure 4 As shown, the key pump acoustic and vibration condition monitoring and intelligent diagnosis system and method includes: I. Overall System Solution.

[0211] This system comprises a complete acoustic and vibration monitoring and intelligent diagnostic system for critical hazardous chemical pumps, consisting of a task control center, an explosion-proof inspection robot, a database of acoustic inspection points for key pumps, an acoustic acquisition module, a visual recognition module, a data analysis engine, and a diagnostic results database. The core idea of ​​the system is as follows: the task control center determines the target pump and the inspection task; the explosion-proof robot moves to the target sampling station according to a safe path; the target pump's identity and relative location are confirmed through visual recognition; and background sound field signals, robot noise templates, and acoustic signals from the target pump are collected in a low-noise parking state. The status of the critical pump is then determined through sound field compensation, baseline matching, and acoustic and vibration characteristic analysis.

[0212] This solution does not rely on manual entry into hazardous areas, nor does it require sensors to be permanently installed on every pump. The robot can perform high-frequency inspections according to scheduled inspection tasks, or it can go to a designated pump for verification and diagnosis after being triggered by fixed monitoring systems, DCS operating data, remote manual commands, or abnormal alarms.

[0213] II. System composition and functions are shown in Table 3 below.

[0214] Table 3. System Composition and Functions

[0215] The navigation path planning module, motion control module, visual recognition module, acoustic acquisition module, and sampling posture adjustment mechanism are all located inside the explosion-proof inspection robot, serving as the robot's local execution and perception modules. The task control center is responsible for issuing target pumps, inspection tasks, sampling stations, and sampling strategies. The robot's internal modules complete path planning, motion control, visual recognition, sampling posture adjustment, and acoustic acquisition based on the task parameters issued by the task control center.

[0216] Furthermore, the internal modules of the robot can interact through the robot's local control unit or local communication bus, without requiring all internal actions to be relayed through the task control center. For example, after the motion control module determines that the robot has reached the target sampling station, it can directly report its arrival status to the local control unit, which then triggers the vision recognition module to collect images of the target area. At the same time, the robot can also send its arrival status back to the task control center, which can record the status or issue a verification command.

[0217] Furthermore, the key pump acoustic inspection point library, data analysis engine, and diagnostic result database are preferably deployed inside the mission control center, or deployed on an edge server / backend server that communicates with the mission control center, and are uniformly called by the mission control center; the explosion-proof inspection robot can cache necessary point parameters, sampling strategies, and temporary sampling data locally, but the complete point library, diagnostic model, and historical diagnostic results are mainly stored on the mission control center side.

[0218] III. Key Pump Acoustic Inspection Point Database

[0219] To ensure the consistency and comparability of the data collected by the robot each time, this system establishes a database of key pump acoustic inspection points. This database records not only the pump's location coordinates but also information related to acoustic sampling, enabling the robot to perform sampling in a stable and repeatable manner.

[0220] The location library can include the following: 1. Key pump and machine serial numbers, names, equipment types, affiliated units, media types, and risk levels; 2. Location coordinates of pumps and motors on the plant area map, accessible routes, hazard level, and obstacle avoidance information; 3. Recommended sampling station location, backup sampling station location, target sampling direction, sampling height, and sampling distance range for the robot; 4. Target detection area information, such as pump body, bearing housing, coupling area, mechanical seal area, inlet pipe section, outlet pipe section, and valve group area; 5. Visual identification information, such as QR codes, equipment nameplates, outline features, pipeline interface locations, or preset visual markers; 6. Historical acoustic baseline data, and operating condition baselines corresponding to speed, flow rate, pressure, temperature, load, valve position, or operating mode; 7. Location information of adjacent pumps, pipelines, valves, compressed air, ventilation equipment, etc., which may form sources of interference.

[0221] IV. Inspection and Sampling Process of Explosion-proof Robots

[0222] When the system is working, the task control center can generate inspection tasks according to a fixed cycle, or it can generate anomaly review tasks based on fixed sensor alarms, DCS operating abnormalities, manual remote commands, or historical trend anomalies. The specific process for the robot to execute tasks is as follows.

[0223] 1. The task control center receives inspection tasks or anomaly verification tasks and determines the key pumps and motors that need to be inspected.

[0224] 2. The task control center queries the key pump acoustic inspection point database for the location coordinates of the target pump, recommended sampling station, backup sampling station, target sampling direction, sampling distance range, historical acoustic baseline and applicable operating condition information.

[0225] 3. The explosion-proof robot plans its path based on the target location, hazardous area information, on-site obstacles, and the robot's current position, and moves to the target sampling station.

[0226] The path planning process requires integrating the information obtained in step 2. The explosion-proof robot uses the recommended sampling station as the priority endpoint and the backup sampling station as the candidate endpoint, taking into account the hazardous area, obstacles, traffic conditions, the robot's current position, battery status, and communication status. Specifically, the task control center sends the robot the target pump location coordinates, recommended sampling station, backup sampling station, target sampling direction, sampling distance range, historical acoustic baseline, and applicable operating conditions to form candidate paths.

[0227] Furthermore, path planning can be performed as follows: First, mark prohibited areas, restricted areas, passable passages, and obstacles on the factory map or navigation grid; then, based on the target pump coordinates, select recommended / backup sampling stations that meet the sampling direction and sampling distance range; next, calculate the path cost from the robot's current position to each candidate sampling station, including path length, hazardous area penalty, obstacle distance penalty, number of turns, communication quality, and sampling posture adjustment amount upon arrival; finally, select the path with the lowest cost and that meets safety constraints as the execution path. If a recommended sampling station is unreachable or safety constraints are not met, automatically switch to a backup sampling station.

[0228] 4. After the robot arrives at the target sampling station, it collects images of the target area through the vision recognition module, identifies QR codes, nameplates, pump body outlines, pipeline interfaces or preset marks, and confirms whether the current detection object is the target pump.

[0229] The appropriate identification method is selected based on the on-site markings and the type of target location. QR codes or barcodes can be identified using decoding; nameplates can be identified using image positioning and OCR; pump body contours, pipeline interfaces, valve group locations, or preset markings can be identified using methods such as target detection, contour matching, key point matching, template matching, or semantic segmentation.

[0230] Furthermore, the confirmation of the current detection target involves comparing the visual recognition results with the corresponding information in the key pump acoustic inspection point database. The comparison includes checking the equipment number, QR code / nameplate information, preset visual markers, pump body contour features, pipeline interface location, target orientation, camera angle, and sampling station location for consistency. When the matching degree between the recognition result and the target pump information in the point database reaches a preset threshold, the system confirms the current detection target as the target pump; if the matching degree is insufficient, the robot readjusts the perspective, acquires images, or returns to the task control center for verification.

[0231] 5. The robot corrects its orientation, sampling distance, and sampling direction based on the visual recognition results, and adjusts the position and orientation of the acoustic acquisition module through the vehicle body, gimbal, lifting mechanism, telescopic mechanism, or robotic arm.

[0232] 6. The robot enters a low-noise parking sampling state. In this state, the robot chassis stops moving, non-essential actuators are paused or operate with reduced noise, and the sampling posture remains stable.

[0233] The prerequisites for the robot to enter the low-noise parking sampling state include: visual recognition confirming the current detection object as the target pump; the robot reaching the target sampling station or backup sampling station; the sampling direction, sampling distance, and sampling height meeting the corresponding requirements in the point database; and the chassis speed, vehicle posture, body noise, communication status, and sampling posture stability all meeting preset conditions. If the target pump's identity is not confirmed or the sampling posture does not meet the requirements, the robot will not enter the effective sampling state but will first perform perspective adjustment, posture correction, or repositioning.

[0234] 7. The acoustic acquisition module acquires background sound field signals, robot body noise templates, and target pump acoustic signals.

[0235] 8. The system determines the reliability of the sampled data; if the reliability is insufficient, the robot adjusts its sampling posture or moves to a backup sampling station for resampling.

[0236] The reliability of the sampled data can be assessed using a tiered approach. A basic sampling gating module is installed within the robot to locally determine fundamental conditions such as robot stability, compliance of sampling distance, excessive body noise, and whether visual recognition confidence meets requirements. The task control center or its data analysis engine is responsible for performing a complete reliability calculation, including background sound field quality, robot body noise template, target sound signal-to-noise ratio, multi-station consistency, operational stability, and historical baseline matching.

[0237] Furthermore, the robot's local module is primarily responsible for rapid gating and immediate resampling, while the task control center is mainly responsible for comprehensive reliability judgment and diagnostic decisions. When the robot's local gating judgment clearly fails to meet the conditions, the sampling posture can be directly adjusted or a backup station can be switched. When the basic conditions are met but the comprehensive reliability is insufficient, the data analysis engine provides feedback on resampling suggestions to the task control center, which then triggers resampling by the task control center or the robot's local control unit.

[0238] 9. After the data collection is completed, the robot uploads the data to the data analysis engine; the data analysis engine performs sound field compensation, working condition baseline matching, sound and vibration feature extraction, and fault diagnosis.

[0239] 10. The diagnostic results are written into the diagnostic results database and generated into tiered alarms, inspection records, diagnostic reports, or maintenance work orders.

[0240] V. Acoustic signal acquisition methods.

[0241] When the robot performs acoustic sampling near the target pump, it does not simply record the on-site sound, but collects multiple types of signals according to fixed position, fixed direction, fixed distance range and stable parking conditions to improve the comparability of data and diagnostic reliability, as shown in Table 4 below.

[0242] Table 4. Description of different acoustic signals

[0243] Furthermore, during the data collection process, the robot needs to determine whether preset parking conditions are met. These preset parking conditions may include the robot chassis speed being zero, the vehicle's posture being stable, the sampling distance being within a preset range, the orientation deviation of the acoustic acquisition module being less than a preset angle, the body noise being lower than a preset threshold, and the visual recognition confidence level meeting requirements. Only when the sampling conditions are met will the system use the collected data as valid acoustic and vibration diagnostic data.

[0244] VI. Multi-site re-sampling and sound source attribution.

[0245] In hazardous chemical installation areas, multiple pumps, compressors, fans, valves, pipelines, and ventilation equipment often operate simultaneously. Sound collected from a single point may not entirely originate from the target pump. To reduce false alarms, this system can be configured with a multi-station re-sampling mechanism.

[0246] When there are adjacent moving devices around the target pump, strong background noise, sampling reliability below the threshold, or uncertainty in the preliminary diagnosis results, the robot moves to a second sampling station or a backup sampling station to collect data on the same target pump again. The system compares the changes in sound energy, spectrum consistency, target direction consistency, distance attenuation relationship, and operating condition stability at different stations to determine whether the abnormal sound source belongs to the target pump.

[0247] For example, if the robot collects abnormal frequency band energy at a station close to the target pump, while the abnormal energy is significantly enhanced at a station far from the target pump but close to an adjacent pump, the system can determine that the abnormality is more likely to come from the adjacent equipment; if multiple stations show that the abnormal acoustic characteristics attenuate steadily with the distance from the target pump and are consistent with the changes in the operating conditions of the target pump, the system can determine that the abnormality is more likely to belong to the target pump.

[0248] VII. Sound Field Compensation and Data Processing.

[0249] After receiving the acoustic data uploaded by the robot, the data analysis engine first performs data parsing, time synchronization, format conversion, abnormal segment removal, and data cleaning. Then, it performs compensation processing on the target pump acoustic signal based on the sampling station position, sampling direction, background sound field signal, and robot body noise template.

[0250] Furthermore, the basic process of sound field compensation processing includes: 1. Estimate the current ambient noise level based on the background sound field signal to reduce the impact of adjacent equipment, pipeline reflections, airflow noise, and ambient noise on the target signal; 2. Based on the robot body noise template, subtract or suppress interference introduced by the robot chassis, motor, fan, power module and structural resonance; 3. Standardize the data from different inspection cycles based on sampling posture, sampling distance, and sampling direction; 4. Based on the sampling results from multiple stations, determine whether the abnormal sound source comes from the target pump, and mark or trigger re-sampling for data with insufficient credibility.

[0251] Furthermore, the compensated acoustic and vibration response characteristics can include sound pressure level changes, root mean square value, peak value, impact characteristics, frequency band energy, dominant frequency component, octave relationship, envelope characteristics, spectral peak changes, spectral centroid, proportion of abnormal frequency band energy, and time trend changes. These characteristics are primarily used to describe the acoustic state of pump operation, and no single formula is used as the sole basis for system diagnosis.

[0252] VIII. Sound and vibration data modeling and diagnostic calculation formulas.

[0253] To ensure that the sound data collected by the robot can more stably characterize the vibration state of the pump, the system does not directly use the original recordings as the diagnostic basis. Instead, it establishes a joint calculation process based on the target pump sound source, the ambient background sound field, the robot's own noise, the sampling posture, and the current operating conditions. The following formulas describe the data processing logic of this system; the symbols can be adjusted according to subsequent engineering implementation.

[0254] set up Indicates the target key pump; Indicates the robot's sampling station location; Indicates the sampling window; Represents discrete sampling points; Indicates the frequency point. Then the robot at the [frequency point]. Raw acoustic sequences collected from each station It can be represented as the superposition of the target pump sound source, background sound field, robot body noise and random disturbance, as shown in the above relationship (2).

[0255] Furthermore, to reflect the influence of the robot's sampling posture on the acoustic signal, the system couples the target sound source with weights. It is related to the sampling distance, sampling direction and attitude deviation, as shown in the above relationship (3).

[0256] Furthermore, before each sampling, the system calculates the parking sampling threshold. Only when the robot's speed, posture, sampling distance, body noise, and visual recognition confidence meet the requirements will the current window be used as a valid sampling window, as shown in the above relation (4).

[0257] Furthermore, the system performs short-time frequency domain transformation on the original signal, background sound field, and robot body noise, respectively obtaining... , , The compensated target acoustic and vibration response can be calculated according to the above relationships (5) and (6).

[0258] Furthermore, after compensation, the system is configured according to the fault-sensitive frequency band set. Extract acoustic vibration energy characteristics. Further, The frequency can be set according to the rotation frequency, frequency multiplication, bearing characteristic frequency band, coupling abnormal frequency band, cavitation high-frequency impact frequency band or empirical frequency band, as shown in the above relationship (9).

[0259] Furthermore, during multi-station re-sampling, the system utilizes the energy attenuation relationship at different stations to determine whether an abnormal sound source belongs to the target pump. For candidate sound sources... Establish sound source attribution residuals As shown in equation (7) above. Simultaneously, the abnormal sound source is assigned to the sound source. probability score As shown in the above relation (8).

[0260] Furthermore, considering that the normal acoustic state of the pump varies under different speeds, loads, flow rates, pressures, temperatures, and valve positions, the system uses operating condition vectors. This indicates the current operating conditions and performs weighted matching in the historical baseline samples, as shown in the above relations (11) to (14).

[0261] Furthermore, the system compares the current compensated acoustic and vibration characteristics with the historical baseline under the matching working condition to obtain the normalized abnormal deviation of the working condition, as shown in the above relationship (15).

[0262] Furthermore, in order to avoid low-quality sampling directly triggering error alarms, the system calculates the sampling reliability based on visual recognition, parking status, sound signal-to-noise ratio, multi-station attribution score and operating condition stability, as shown in the above relationship (4).

[0263] Ultimately, the system can construct an acoustic and vibration anomaly index for critical pumps and machinery used in hazardous chemicals. Furthermore, the state is classified in conjunction with the trend term, as shown in the above relations (16) and (17). It can be used to output status levels such as normal, attention, abnormal, and severe abnormality, and can also be used as the basis for calculation to generate re-mining tasks, graded alarms, and maintenance suggestions.

[0264] IX. Baseline Matching and Intelligent Diagnosis.

[0265] The operating sound of key pumps varies under different speeds, loads, flow rates, pressures, temperatures, valve positions, and media conditions. Directly comparing the sounds under different operating conditions can easily lead to misinterpreting normal operating changes as malfunctions. Therefore, this system incorporates an operating condition baseline matching mechanism during diagnostics, as shown in Table 2 above.

[0266] The system acquires the current operating condition information of the target pump and selects baseline data that matches the current operating condition from historical acoustic baselines. During diagnosis, the current acoustic and vibration response characteristics are compared with historical baselines under the corresponding operating conditions, baselines of similar equipment, or data from the previous inspection cycle to calculate the degree of abnormal deviation. Then, the pump status is output in conjunction with the fault model.

[0267] The diagnostic results can include status levels such as normal, noteworthy, abnormal, and severely abnormal. They can also further output the suspected fault type, suspected fault location, abnormal trend, risk level, and suggested verification method. For data that does not meet the sampling reliability requirements, the system does not directly output a definitive fault conclusion, but instead marks it as requiring re-sampling or remote manual confirmation.

[0268] 10. The setup of the robotic arm.

[0269] A robotic arm can be included in this solution, but it is not the only necessary condition for the system to operate. The robotic arm serves as an optional implementation of the sampling posture adjustment mechanism, used to further adjust the position of the acoustic acquisition module or auxiliary sensors relative to the target detection area after the robot has reached the target sampling station.

[0270] In embodiments without a robotic arm, the robot can adjust the orientation, height, and distance of the acoustic acquisition module through vehicle steering, gimbal rotation, lifting mechanisms, or telescopic mechanisms to achieve non-contact near-field acoustic acquisition. In embodiments with a robotic arm, the robotic arm can move the acoustic acquisition module to a position closer to the pump body, bearing housing, coupling area, mechanical seal area, inlet pipe section, outlet pipe section, or valve group area based on the target detection location determined by the visual recognition module, in order to obtain acoustic signals closer to the target location.

[0271] The robotic arm is primarily used for proximity sampling, orientation adjustment, or auxiliary detection; it is not required to perform disassembly, repair, sealing, or other high-risk operations in the main process. If a contact-type auxiliary sensor is added to the subsequent system, the robotic arm can also briefly approach or contact the target detection area for supplementary sampling, provided that remote authorization and safety conditions are met.

[0272] XI. Anomaly Handling and Data Closure.

[0273] When the system determines that a critical pump or motor is malfunctioning, the diagnostic results database stores the target pump or motor number, sampling time, sampling location, visual recognition results, operating condition information, raw acoustic data, characteristics after sound field compensation, diagnostic conclusion, risk level, and handling records. The task control center generates different handling actions based on the risk level.

[0274] 1. Low-risk anomalies: Generate a record of concern and automatically review it in the next inspection cycle; 2. Medium-risk anomaly: Generate a tiered alarm, prompting remote personnel to view images, acoustic data, and trend curves, and assign maintenance personnel to pay close attention; 3. High-risk anomalies: Trigger re-inspection tasks, maintenance work orders, or remote confirmation processes, and if necessary, link fixed monitoring systems and process systems for further verification; 4. Insufficient sampling reliability: The robot adjusts its sampling posture, switches to a backup station, or extends the sampling time, and then re-uploads the sampling data.

[0275] Through the aforementioned closed loop, the system can connect "inspection task - robot arrival - visual confirmation - acoustic acquisition - data diagnosis - risk classification - review and handling - result archiving" to form an unmanned inspection and intelligent diagnosis process for critical pumps and machinery in hazardous chemicals.

[0276] 12. Example of specific implementation process.

[0277] Implementation Process 1: Scheduled Inspections. The system generates key pump inspection tasks according to a preset inspection cycle. The explosion-proof robot moves to the vicinity of the target pump according to the path issued by the task control center, confirms the pump's identity through visual recognition, and collects background sound field, body noise template, and target pump acoustic signals in a low-noise parking state. After the data is uploaded, the data analysis engine combines historical acoustic baselines and current operating conditions to determine the pump's status and generates an inspection record.

[0278] Implementation Process Two: Anomaly Verification. When the fixed monitoring system or DCS data detects an abnormal trend in temperature, pressure, flow, or vibration of a critical pump, the task control center generates an anomaly verification task. The explosion-proof robot first proceeds to the sampling station corresponding to the pump for acoustic verification. If the sampling reliability is insufficient or there is significant interference from adjacent equipment on site, the robot automatically switches to a backup station to collect data from multiple stations, and then the system determines whether the anomaly belongs to the target pump.

[0279] Implementation Step 3: Configuring the Robotic Arm for Auxiliary Sampling. After the explosion-proof robot reaches the target pump, the vision recognition module determines the location of the target detection area. The control system controls the robotic arm to move the acoustic acquisition module to an auxiliary sampling pose close to the target detection area, and acquires near-field acoustic signals after the robotic arm stabilizes. This method is used to improve the consistency of sampling distance and sampling angle.

[0280] Implementation Process Four: Fault Trend Tracking. For the same critical pump, the system continuously saves the acoustic and vibration response characteristics of multiple inspection cycles and stores them according to operating conditions. When the acoustic and vibration characteristics under the same operating condition continuously deviate from the historical baseline, the system uses this deviation trend as an early anomaly indicator and generates alerts for attention, review, or maintenance.

[0281] Figure 5This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0282] The following is a detailed reference. Figure 5 This diagram illustrates a suitable structural schematic for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 501, which can perform various appropriate actions and processes based on a program stored in read-only memory (ROM) 502 or a program loaded from memory 508 into random access memory (RAM) 503. RAM 603 also stores various programs and data required for the operation of the electronic device. The processor 501, ROM 502, and RAM 503 are interconnected via bus 504. An input / output (I / O) interface 506 is also connected to bus 504.

[0283] Typically, the following devices can be connected to I / O interface 506: input devices 506 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 507 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 508 including, for example, magnetic tapes, hard disks, etc.; and communication devices 509. Communication device 509 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 5 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0284] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 509, or installed from a memory 508, or installed from a ROM 502. When the computer program is executed by the processor 501, it performs the functions defined in the critical pump fault diagnosis method of the embodiments of the present invention.

[0285] Figure 5 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0286] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the key pump fault diagnosis method shown in the above embodiments is implemented.

[0287] A portion of this invention can be applied as a computer program product, such as computer program instructions, which, when executed by a computer, can invoke or provide the methods and / or technical solutions according to the invention through the operation of the computer. Those skilled in the art will understand that the forms in which computer program instructions exist in a computer-readable medium include, but are not limited to, source files, executable files, installation package files, etc. Correspondingly, the ways in which computer program instructions are executed by a computer include, but are not limited to: the computer directly executing the instructions, or the computer compiling the instructions and then executing the corresponding compiled program, or the computer reading and executing the instructions, or the computer reading and installing the instructions and then executing the corresponding installed program. Here, the computer-readable medium can be any available computer-readable storage medium or communication medium accessible to a computer.

[0288] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A fault diagnosis system for key pumps, characterized in that, The system includes a task control center and an explosion-proof inspection robot. The task control center includes a database of acoustic inspection points for key pumps and motors. The explosion-proof inspection robot includes a visual recognition module, an acoustic acquisition module, and a local control unit. The task control center is used to generate inspection tasks according to a preset fixed cycle, determine the target key pumps according to the inspection tasks, obtain the location information set of the target key pumps from the acoustic inspection point database of the key pumps, and send the location information set and the acoustic inspection point database of the key pumps to the explosion-proof inspection robot. The explosion-proof inspection robot is used to move to the first target sampling station based on the point information set, and based on the key pump acoustic inspection point library, to acquire a complete set of sampling datasets that meet the local basic reliability requirements under low-noise parking sampling state through the visual recognition module and acoustic acquisition module, and send the complete set of sampling datasets to the task control center. The low-noise parking sampling state meets multi-dimensional joint gating conditions, and the complete set of sampling datasets includes background sound field signals, robot body noise templates, acoustic signals of target key pumps, and auxiliary state datasets. The task control center is also used to perform data processing, sound field compensation, adaptive noise suppression, multi-station sound source attribution discrimination, diagnostic calculation and working condition baseline matching on the complete set of sampling datasets in sequence, and to obtain the target fault diagnosis results of the target key pump. The visual recognition module is used to acquire a corresponding target area image based on a visual acquisition command, recognize the target area image to obtain a visual recognition result, and send the visual recognition result to the local control unit. The local control unit is used to determine whether the detection object contained in the target area image is the target key pump based on the visual recognition result and the key pump acoustic inspection point library, and when the detection object is the target key pump, to control the explosion-proof inspection robot to be in a low-noise parking sampling state based on preset parking conditions, and to send an acoustic acquisition command to the acoustic acquisition module. The acoustic acquisition module is used to acquire the background sound field signal, the robot body noise template, the acoustic signal of the target key pump, and the auxiliary status dataset based on the acoustic acquisition command, and send the background sound field signal, the robot body noise template, the acoustic signal, and the auxiliary status dataset to the local control unit; The local control unit is further configured to encapsulate the background sound field signal, the robot body noise template, the acoustic signal and the auxiliary state dataset to obtain the complete set of sampling datasets, and to perform local basic credibility judgment on the complete set of sampling datasets, and when the local basic credibility meets the first preset threshold requirement, to send the complete set of sampling datasets to the task control center.

2. The system according to claim 1, characterized in that, The explosion-proof inspection robot also includes: a navigation path planning module, a movement control module, and a local control unit; The navigation path planning module is used to plan a path based on the point information set, obtain the target movement path, and send the target movement path to the movement control module through the local control unit. The mobile control module is used to control the movement of the explosion-proof inspection robot based on the target movement path, and when the explosion-proof inspection robot moves to the first target sampling station, it sends a movement completion command to the local control unit, so that the local control unit sends a visual acquisition command to the visual recognition module based on the movement completion command.

3. The system according to claim 1, characterized in that, The explosion-proof inspection robot also includes: a sampling posture adjustment mechanism; The local control unit is also used to send a first adjustment command to the sampling posture adjustment mechanism when the detected object is not the target key pump; The sampling posture adjustment mechanism is used to adjust the position and orientation of the explosion-proof inspection robot based on the first adjustment command, and send a first adjustment completion command to the local control unit according to the adjusted position and orientation, so that the local control unit sends a new visual acquisition command to the visual recognition module based on the first adjustment completion command.

4. The system according to claim 3, characterized in that, The local control unit is further configured to send a second adjustment command to the sampling posture adjustment mechanism when the local basic confidence level does not meet the first preset threshold requirement. The sampling posture adjustment mechanism is further configured to adjust the position and orientation of the explosion-proof inspection robot based on the second adjustment command, and send a second adjustment completion command to the local control unit according to the adjusted position and orientation, so that the local control unit sends a new acoustic acquisition command to the acoustic acquisition module based on the second adjustment completion command.

5. The system according to claim 2, characterized in that, The local control unit is also used to send a movement command to the movement control module when the local basic trust level does not meet the first preset threshold requirement, so that the movement control module controls the explosion-proof inspection robot to move to the second target sampling station or the backup sampling station based on the movement command.

6. The system according to claim 1, characterized in that, The task control center includes: a data analysis engine and a diagnostic results database; The data analysis engine is used to perform a comprehensive confidence assessment of the entire set of sampled datasets across all dimensions. When the comprehensive confidence assessment across all dimensions meets the second preset threshold requirement, the engine sequentially performs data processing, sound field compensation, adaptive noise suppression, multi-station sound source attribution discrimination, diagnostic calculation, and working condition baseline matching on the entire set of sampled datasets, and obtains the target fault diagnosis result of the target key pump. The diagnostic results database is used to store the diagnostic results of the target fault.

7. The system according to claim 1, characterized in that, The task control center is also used to generate anomaly review tasks based on fixed sensor alarms, DCS operating condition anomalies, manual remote commands, or historical trend anomalies, and to determine the target key pumps based on the anomaly review tasks. The task control center is also used to acquire historical acoustic and vibration characteristics that continuously deviate from the baseline, and to generate fault trend tracking and diagnosis tasks based on the operating conditions.

8. A method for diagnosing faults in key pumps, characterized in that, A task control center in a critical pump fault diagnosis system according to any one of claims 1 to 7; the method includes: Receive the complete set of sampled data of the target key pump sent by the explosion-proof inspection robot in the key pump fault diagnosis system; Data processing, sound field compensation, adaptive noise suppression, and multi-station sound source attribution discrimination are performed on the complete set of sampling datasets to obtain the compensated acoustic-vibration response feature set; The acoustic vibration response feature set is subjected to index extraction, and an acoustic vibration feature vector set is constructed. Based on the key pump acoustic inspection point library, a real-time operating condition vector is constructed, and based on the real-time operating condition vector and the acoustic vibration feature vector, a comprehensive acoustic vibration anomaly index is constructed. Based on the comprehensive acoustic and vibration anomaly index, the foundation condition level of the target key pumps is determined; Based on the comprehensive acoustic and vibration anomaly index, the acoustic and vibration feature vector set, the real-time operating condition vector, and the basic state level, fault diagnosis is performed using preset fault type acoustic and vibration judgment rules to obtain the target fault diagnosis result of the target key pump.

9. The method according to claim 8, characterized in that, Based on the real-time operating condition vector and the acoustic vibration feature vector, a comprehensive acoustic vibration anomaly index is constructed, including: Based on the real-time operating condition vector, determine the mean and covariance matrix of the normal acoustic and vibration baseline that matches the operating conditions of the target key pump; Based on the acoustic vibration feature vector set, the mean of the normal acoustic vibration baseline and the covariance matrix, the normalized abnormal deviation of the working condition is determined. The comprehensive acoustic and vibration anomaly index is constructed based on the normalized anomaly deviation of the operating conditions.

10. The method according to claim 8, characterized in that, The method further includes: Based on the target fault diagnosis results, a graded closed-loop handling plan is generated.

Images