Intelligent operation and maintenance management control system for full life cycle of fire-fighting equipment

Abstract

The invention belongs to the technical field of fire-fighting facility control and management, and discloses a fire-fighting facility full-life-cycle intelligent operation and maintenance management control system. Comprising the following steps: acquiring real-time field state data; in a non-fire state, executing an active excitation and response test to obtain dynamic response data; processing the field state data and the dynamic response data by adopting an end-to-end multi-task learning framework, and outputting a structured diagnosis result containing the health states of the pump set, the valve and the pipe network; establishing an initial health baseline of each key device, and dynamically generating a self-adaptive alarm threshold value; a quantitative risk level is generated in combination with the fire safety level, and an operation and maintenance strategy instruction containing pilot operation and maintenance parameters is generated; main and standby equipment switching and strong health detection are automatically executed, and the fire protection function is preferentially guaranteed when a fire alarm or a manual forcing instruction is received; and executing grading alarm according to the quantitative risk grade. And intelligent operation and maintenance management control of the whole life cycle of fire-fighting equipment is realized.

Description

Technical Field

[0001] This invention relates to the field of fire protection facility control and management technology, and more specifically, to an intelligent operation and maintenance management and control system for the entire life cycle of fire protection facilities. Background Technology

[0002] In modern industrial sectors, especially in processing workshops with continuous production characteristics and warehouses storing large quantities of materials, fire protection facilities serve as the last line of defense, directly impacting the safety of people and property. The core requirement for fire protection facilities in such environments is constant readiness; that is, they must be in a state of immediate and operational readiness throughout their entire life cycle. Failure of any facility, especially during a sudden fire, can lead to catastrophic consequences.

[0003] Traditional operation and maintenance management mainly relies on periodic manual inspections and planned maintenance. However, manual inspections are discrete point-in-time activities, unable to achieve continuous, real-time perception of equipment status. During the intervals between inspections, equipment deterioration cannot be detected in time, leaving facilities in a long-term, undetected risk blind spot. Therefore, using intelligent detection is the current trend in the operation and maintenance management of fire protection facilities. However, existing systems monitor mostly operational result parameters (such as whether the pressure at the end of the pipeline meets the standard), rather than the health status parameters of the fire protection equipment itself (such as pump shaft wear and vibration characteristics, the degree of aging of motor winding insulation, and the torque change curve of valve actuators). Therefore, these hidden faults, such as internal corrosion of fire pumps, false opening or false closing of key electric or pneumatic valves (i.e., the actuator feedback signal is normal but the valve core is not actually in place), local blockage inside the pipeline, and slow aging of seals, cannot be detected through visual inspection or simple functional start-stop tests. These problems are usually hidden, but they can cause sudden shutdowns or malfunctions of fire protection facilities during emergency activation, rendering the entire fire protection system ineffective. Therefore, how to use advanced technologies in the field of automatic control to achieve real-time online monitoring and early warning of hidden faults such as pump corrosion and valve false opening, and prevent sudden shutdowns, has become a unique technical problem in this field that urgently needs to be solved but has not yet been well resolved.

[0004] In view of the above-described problems, this design provides an intelligent operation and maintenance management and control system for the entire life cycle of fire protection facilities. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art and to achieve the above objectives, the present invention provides the following technical solution: an intelligent operation and maintenance management and control system for the entire life cycle of fire protection facilities, comprising: The data acquisition and active response testing module is used to collect the operating status of fire pumps, motors, pipelines and valve actuators through multi-source sensing to obtain real-time field status data; and in non-fire conditions, if the main fire pipeline meets the pressure conditions, it performs active excitation and response testing to obtain dynamic response data. The health diagnosis and latent fault identification module adopts an end-to-end multi-task learning framework to process field status data and dynamic response data, and outputs structured diagnostic results including the health status of pump sets, valves and pipelines. The full lifecycle management module is used to establish the initial health baseline of each key device, construct a historical sequence of health indicators based on structured diagnostic results, fit the device aging trajectory function, and dynamically generate adaptive alarm thresholds. The risk assessment and strategy generation module is used to generate a quantitative risk level by combining the fire safety level, and to generate operation and maintenance strategy instructions that include trial operation and maintenance parameters. The intelligent linkage control execution module is used to convert operation and maintenance strategy instructions into control instructions, automatically execute equipment master / slave switching and strong health detection, and prioritize the protection of fire protection functions when a fire alarm or manual forced command is received. The adaptive optimization and alarm module is used to adaptively adjust the data acquisition strategy according to the stability of the equipment status and execute graded alarms according to the quantified risk level.

[0006] Preferably, the method for multi-source sensing acquisition of the operating status of fire pumps, motors, pipelines, and valve actuators includes: Multiple acceleration sensors are installed on the bearing housing and pump body of the fire pump to collect vibration signals of the bearing and pump body. The three-phase current, voltage, and active power waveforms of the fire pump motor are collected using electrical parameter sensors. Pressure sensors and flow meters are installed at key nodes of the fire protection pipeline network to collect pressure and flow data of the network. The limit switch status, encoder position signal, and execution torque curve of the electric valve actuator are collected by position and torque sensors.

[0007] Preferably, the method for performing active stimulus and response testing includes: Determine if the pressure in the main fire hydrant pipeline is not lower than the preset lower limit; if so, then... Controlled flow or pressure disturbances are injected into the fire pipeline by briefly adjusting the speed of the fire pump or briefly controlling the opening of the bypass drain valve. Perform short-stroke reciprocating opening and closing actions at a preset angle or apply short-stroke torque loading to key valves; During the active incentive execution, multi-source sensor data is collected to obtain dynamic response data; When the pressure in the main fire pipeline is detected to be lower than the preset lower limit, the active excitation and response test will be stopped. Furthermore, real-time on-site status data refers to data acquired through multi-source sensing during the full lifecycle management of fire protection facilities.

[0008] Preferably, the end-to-end multi-task learning framework includes a shared input layer, one shared encoder, and three dedicated task decoders. The shared input layer is used to receive field status data and dynamic response data. The three dedicated task decoders include a pump health head, a valve identification head, and a pipeline diagnostic head. The pump health head is designed based on a multilayer perceptron, including an input layer, hidden layers, and an output layer. The input layer receives the output of the shared encoder. The hidden layers consist of 2 to 4 fully connected layers, with the number of neurons decreasing progressively in each layer. The hidden layers are activated using the ReLU function. The output layer outputs the bearing wear level and the degree of pump corrosion. If the bearing wear level is predefined as N, then the number of neurons in the output layer is N+1. The Softmax function is used to activate the bearing wear level task, outputting the probability of each level; the Sigmoid function is used to activate the pump corrosion degree task, outputting a value between 0 and 1. The valve recognition head is designed based on a contrastive learning structure, including a two-branch Siamese network. Each branch is a 3-4 layer multilayer perceptron, and the two branches share weights. The input of branch one is the valve's stroke feedback signal features; the input of branch two is the pressure-flow response features of the pipeline network within the same time period. The two sub-networks output vectors of the same dimension. Then, the distance between the two branch vectors is calculated. This distance is converted into an inconsistency confidence score with a value range of 0-1 through a fully connected layer and a sigmoid function. The pipeline diagnostic head is designed based on a graph neural network decoder. First, the fire pipeline network is abstracted as a graph, where each node represents the physical location of a pipeline point. Node features include static attributes and dynamic data, and edges represent pipeline connections. Edge features include pipeline scalar values. A 2-4 layer graph convolutional network is deployed, with the number of layers related to the diameter of the pipeline topology. After graph convolution, each node obtains a feature vector containing global topology information. At this point, a fully connected layer is connected to each node, outputting a scalar value for each node. After processing by the sigmoid function, this value is transformed into the fault confidence of that node. The output is a probability list, with the list length equal to the number of nodes. Each value in the list corresponds to the fault confidence of a node. The multi-task learning framework generates a structured diagnostic result by aggregating the outputs of three task decoders. This result includes pump group health status assessment results, valve consistency identification results, and a list of pipeline node fault confidence.

[0009] Preferably, the method for dynamically generating adaptive alarm thresholds includes: Quantitative features of the health status of each key device are extracted from the structured diagnostic results and used as core health indicators. In the initial stage of fire protection facility operation, core health indicators under multiple working conditions are collected to establish the initial health baseline vector of each key equipment. During equipment operation, core health indicators are recorded in chronological order to form a historical sequence of health indicators. Based on the historical sequence of health indicators, regression analysis is used to fit the aging trend of equipment health status with operating time, and the equipment aging trajectory function is obtained. Reversible operating condition fluctuations are extracted from real-time health indicators, and corrected health indicators that reflect the inherent performance degradation of the equipment are output. Based on the equipment aging trajectory function, the corresponding adaptive alarm threshold is dynamically calculated and updated for each core health indicator.

[0010] Preferably, the method for separating reversible operating condition fluctuations from real-time health indicators and outputting corrected health indicators that reflect the inherent performance degradation of the equipment includes: Establish a lookup table for the impact of operating conditions, with season, water source, and ambient temperature and humidity as inputs and typical deviations of health indicators as outputs; The offset of the current operating condition is obtained by looking up the operating condition influence lookup table, and the real-time core health indicators are corrected. This is recorded as the corrected health indicator after removing reversible operating condition fluctuations.

[0011] Preferably, the method for generating a quantitative risk level by combining fire safety level with the generation of operation and maintenance strategy instructions that include trial operation and maintenance parameters includes: Obtain the aging rate represented by the device aging trajectory function; Based on the current fire safety level requirements, the corrected health indicators are compared with the adaptive alarm threshold, and the aging rate is evaluated, thereby quantifying the failure risk of individual equipment and the overall fire protection system and outputting the quantitative risk level. Based on the quantified risk level, the pre-stored operation and maintenance strategy library is queried to generate corresponding operation and maintenance strategy instructions. The operation and maintenance strategy instructions shall at least include the trial operation configuration parameters for fire pumps and the maintenance suggestion parameters for equipment with identified hidden faults.

[0012] Preferably, the method for converting operation and maintenance policy instructions into control instructions includes: During non-fire periods, the fire pumps are automatically put into trial operation according to the trial operation configuration parameters, and active excitation and response tests are triggered simultaneously during the trial operation. When the corrected health indicators are lower than their corresponding adaptive alarm thresholds, or the quantified risk level reaches the preset high-risk level, the main and backup equipment are automatically switched, and enhanced health detection is performed on the switched main equipment. Upon receiving a fire alarm signal or a manual forced pump start command, all active excitation tests shall be immediately stopped.

[0013] Preferably, the method for adaptively adjusting the data acquisition strategy based on the stability of the equipment status includes: when the change range of the monitoring values ​​corresponding to the core health indicators for multiple consecutive periods is less than a first preset threshold, it is determined that the fire protection facilities are stable, and data acquisition is carried out at a low sampling rate. When the amplitude or current value of the vibration signal collected in real time exceeds the second preset threshold, an abnormal sign is determined, the sampling rate is automatically increased, and an active excitation test is triggered.

[0014] Preferably, the method for executing graded alarms based on quantified risk levels includes: Low-level risks are pushed to maintenance work orders, medium-level risks are notified to on-duty personnel via mobile messages, and high-level risks trigger fire alarm signals and are uploaded to the superior management platform.

[0015] The technical effects and advantages of the intelligent operation and maintenance management and control system for the entire life cycle of fire protection facilities according to the present invention are as follows: By combining proactive stimulation, multi-source sensing, and multi-task fusion, online, real-time, and location-based diagnosis of hidden faults can be achieved in the field of fire protection facility operation and maintenance, shifting the operation and maintenance mode from post-fault repair to pre-fault prediction.

[0016] By establishing an initial baseline, aging trajectory, and adaptive threshold based on individual devices, and by removing operational interference, a health record and aging model were created for each device, enabling precise and personalized health assessments and greatly reducing false alarms and missed alarms.

[0017] This design forms a self-operating intelligent operation and maintenance closed loop. It can not only detect problems, but also assess their severity and automatically execute predetermined countermeasures, significantly reducing reliance on human experience and improving response speed and decision consistency. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of an intelligent operation and maintenance management and control system for the entire life cycle of fire protection facilities according to the present invention; Figure 2 This is a schematic diagram of the method for obtaining tire batch distribution characteristics in this invention; Figure 3 This is a schematic diagram of the ternary heterogeneous diagram in this invention. Detailed Implementation

[0019] 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, and 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.

[0020] Please see Figure 1 , Figure 2 and Figure 3 In this embodiment of the invention, a smart operation and maintenance management and control system for the entire life cycle of fire protection facilities includes: The data acquisition and active response testing module is used to collect the operating status of fire pumps, motors, pipelines and valve actuators through multi-source sensing to obtain real-time field status data; and in non-fire conditions, if the main fire pipeline meets the pressure conditions, it performs active excitation and response testing to obtain dynamic response data. The health diagnosis and latent fault identification module adopts an end-to-end multi-task learning framework to process field status data and dynamic response data, and outputs structured diagnostic results including the health status of pump sets, valves and pipelines. The full lifecycle management module is used to establish the initial health baseline of each key device, construct a historical sequence of health indicators based on structured diagnostic results, fit the device aging trajectory function, and dynamically generate adaptive alarm thresholds. The risk assessment and strategy generation module is used to conduct quantitative risk assessments based on fire safety levels and generate operation and maintenance strategy instructions that include trial operation and maintenance parameters. The intelligent linkage control execution module is used to convert operation and maintenance strategy instructions into control instructions, automatically execute equipment master / slave switching and strong health detection, and prioritize the protection of fire protection functions when a fire alarm or manual forced command is received. The adaptive optimization and alarm module is used to adaptively adjust the data acquisition strategy according to the stability of the equipment status and execute graded alarms according to the quantified risk level.

[0021] Addressing the core challenges of traditional technologies, such as the inability of manual inspections to continuously perceive and the failure of traditional monitoring to detect latent faults like pump corrosion and valve false opening, this solution utilizes a multi-source sensor network to directly perceive the internal state of equipment (e.g., bearing vibration, motor winding current, and valve torque). Building upon this, non-invasive active excitation testing (fine-tuning speed and micro-moving valves) injects safe and controllable "test signals" into the system, thereby stimulating and capturing the characteristics of latent faults in dynamic responses. This fundamentally changes the data acquisition method, shifting from passively waiting for faults to manifest to actively detecting and stimulating fault characteristics. This provides a high-quality, high-dimensional dynamic data foundation for subsequent intelligent diagnosis, overcoming the bottleneck of unseen raw data for latent faults. The specific design is as follows: The method for multi-source sensing acquisition of the operating status of fire pumps, motors, pipelines, and valve actuators includes: Multiple acceleration sensors are installed on the bearing housing and pump body of the fire pump to collect vibration signals from the bearings and pump body. These sensors are used to determine the condition of bearing wear and pump body corrosion. The sensors are fixed to the surface of the fire pump bearing housing and pump body using glue or bolts. To achieve vibration monitoring, the installation points should be selected in areas prone to failure, such as bearings. Specific placement points can be as follows: Deployment point location Monitoring targets Installation direction motor drive end bearing housing Motor bearing condition, rotor balance Radial (horizontal or vertical) Motor non-drive end bearing housing Motor bearing condition radial Water pump drive end bearing housing (near coupling) Water pump bearing condition, shaft alignment radial and axial Water pump non-drive end bearing housing Water pump bearing condition radial Pump body outlet section casing Impeller imbalance, cavitation radial The three-phase current, voltage, and active power waveforms of the fire pump motor are collected using electrical parameter sensors to analyze motor load changes and winding insulation degradation characteristics. Specifically, the voltage monitoring terminals of the sensors are directly connected in parallel to the output terminals of the main power circuit breaker of the fire pump motor to collect voltage signals; open-type current transformers are installed on each phase power line, and then the secondary output lines of the transformers are connected to the current monitoring terminals to collect current signals.

[0022] Pressure sensors (e.g., connecting a pressure tap to the pipe and then installing the sensor) and flow meters (e.g., clamping the sensor to the outer wall of the pipe) are installed at key nodes of the fire protection pipe network (referring to locations that are crucial to assessing the overall health and water supply capacity of the system, and whose selection is mainly based on hydraulic analysis and engineering principles) to collect pressure and flow data of the pipe network; this data is used to determine whether there is local blockage in the pipe network and the actual opening degree of the valves. Position and torque sensors (usually integrated within electric or pneumatic actuators) are used to collect the limit switch status, encoder position signal, and actuation torque curve of the electric valve actuator. This data is used to identify valve core jamming and false opening / closing behaviors. The collected data is synchronized with a unified timestamp to ensure the multi-source time sequence alignment of the field status data.

[0023] The method for performing active stimulus and response testing includes: The purpose of active excitation and response testing is to artificially create a transient flow or pressure fluctuation in the pipeline as a test signal, without affecting the main and backup fire water pressure. By collecting dynamic response data of the fire equipment to this signal, faults that cannot be detected under steady-state conditions can be analyzed, such as sluggish pump response (possibly due to internal corrosion) or abnormal pipeline damping (possibly due to partial blockage). Additionally, in non-fire conditions, non-full-stroke, small-amplitude functional tests are performed on important electric valves (such as zone control valves and pump outlet check valves). The aim is to verify the flexibility of their mechanical actuators and whether the actual valve core position matches the feedback signal, thereby diagnosing latent faults such as false opening and closing.

[0024] Determine if the pressure in the main fire hydrant pipeline is not lower than the preset lower limit; if so, then... Controlled flow or pressure disturbances are injected into the fire pipeline by briefly adjusting the fire pump speed or briefly opening the bypass drain valve. Specifically, briefly adjusting the fire pump speed involves controlling the pump's frequency converter to slightly increase its speed from the rated 2950 rpm to 3050 rpm within 3 seconds, maintaining this speed for 2 seconds, and then returning it to 2950 rpm. This operation injects a small additional flow into the pipeline. With a healthy pump and unobstructed pipeline, the pressure rise and fall response curve should be rapid and smooth. If internal corrosion of the pump body leads to decreased efficiency, or if there is localized blockage in the pipeline network, the response will exhibit delays, fluctuations, or failure to reach the expected peak value.

[0025] Short-term control of the bypass drain valve opening refers to rapidly opening a small-diameter test valve installed on the main pipeline bypass to 30% and maintaining it for 5 seconds, then closing it. This operation creates a small, controlled pressure release in the pipeline. The rate of pressure decay and recovery characteristics can be analyzed by acquiring high-frequency pressure waveforms. In a healthy, well-sealed pipeline system, pressure drop and recovery are stable and predictable. However, if there is internal leakage in the valve or poor sealing in the pipeline network, the pressure decay will be abnormally accelerated.

[0026] The system performs preset-angle, small-stroke reciprocating opening and closing actions or applies small-stroke torque loads to key valves. Specifically, the preset-angle, small-stroke reciprocating opening and closing action means that, with the valve currently fully open (90 degrees), it is controlled to first close to 85 degrees, and then open back to 90 degrees, forming a small-stroke reciprocating motion of 5 degrees. This operation aims to stimulate the response of mechanical components such as the valve stem and bearings. The actual stroke curve of the valve stem and the motor drive torque curve are synchronously collected by position and torque sensors. A healthy valve should have a smooth stroke curve and a smooth torque curve with peak values ​​within the normal range. If the valve has valve stem jamming or bearing dry friction, the torque curve will show abnormal spikes; if there is excessive gearbox clearance, the stroke curve will show obvious hysteresis.

[0027] Applying a small-stroke torque load refers to momentarily outputting a short-duration torque pulse (e.g., lasting 0.5 seconds) slightly higher than the normal holding torque while the valve remains stationary at a certain opening degree (e.g., 50 degrees), and then immediately resuming. This operation aims to test whether the valve core is at risk of sticking or seizing. Monitoring the actual position of the valve at the moment the torque pulse is applied reveals any slight movement. If a valve is falsely fixed due to scale or rust from prolonged inactivity, the valve core may not actually rotate under a small torque load, but its limit switch feedback signal may appear normal due to gear idling. This contrast between a normal signal and no valve core movement is key to diagnosing false opening or false closing.

[0028] During the active incentive execution, multi-source sensor data is collected to obtain dynamic response data; When the pressure in the main fire pipeline is detected to be lower than the preset lower limit, the active excitation and response test will be stopped. Furthermore, real-time on-site status data refers to data acquired through multi-source sensing during the full lifecycle management of fire protection facilities.

[0029] To address the limitations of single-parameter threshold alarms in handling complex and coupled latent fault modes, and the difficulties and inefficiencies of multiple independent diagnostic models working together, this paper proposes a fusion model architecture combining a shared encoder and a multi-task decoder. The shared encoder automatically extracts deep common features from multi-source data, and dedicated decoders perform quantitative evaluation, consistency verification, and spatial localization, achieving comprehensive diagnostic results for pumps, valves, and pipelines. This not only significantly improves analysis efficiency but also implicitly learns the correlations between faults in different components through a shared feature layer, making the diagnosis more accurate and reliable. The specific design is as follows: The end-to-end multi-task learning framework includes a shared input layer, one shared encoder, and three dedicated task decoders. The shared input layer receives field status data and dynamic response data. The three dedicated task decoders include a pump health head, a valve identification head, and a pipeline diagnostic head. The pump health head is designed based on a multilayer perceptron, including an input layer, hidden layers, and an output layer. The input layer receives the output of the shared encoder. The hidden layers consist of 2 to 4 fully connected layers, with the number of neurons decreasing layer by layer. For example, with 3 fully connected layers, the first layer would have 1024 neurons, the second 512 neurons, and the third 256 neurons. This decreasing layer structure helps the network to be hierarchical and to selectively combine features, ultimately focusing on the most relevant fault modes. The hidden layers are activated using the ReLU function. The output layer outputs the bearing wear level and the pump body corrosion degree. If the bearing wear level is predefined as N, then the number of neurons in the output layer is N+1. This is because if the bearing wear level is defined as mild, moderate, or severe, then 3 neurons are needed for that level. For the continuous value corresponding to the pump body corrosion degree, 1 neuron is needed to output a normalized exponent between 0 and 1. Use the Softmax function to activate the bearing wear level task and output the probability of each level; use the Sigmoid function to activate the pump body corrosion level task and output a value between 0 and 1. The valve recognition head is designed based on a contrastive learning structure. The core of this structure is comparison and verification, rather than direct classification. It is used to detect inconsistencies between the valve's "claimed position" (feedback signal) and its "actual impact" (pressure / flow response). It includes a two-branch twin network, with each branch being a 3-4 layer multilayer perceptron (e.g., 128→64→32). The two branches share weights. The input to branch one is the valve's stroke feedback signal characteristics; the input to branch two is the pressure-flow response characteristics of the pipeline network within the same time period. The stroke feedback signal characteristics and pressure-flow response characteristics refer to the high-dimensional abstract feature vectors generated by the shared encoder within its internal network layers after analyzing the original sensor data, representing the essence of the signal. The two sub-networks output vectors of the same dimension. Then, the distance between the two branch vectors (e.g., Euclidean distance or cosine distance) is calculated. This distance value is converted into an inconsistency confidence score ranging from 0 to 1 through a fully connected layer and a sigmoid function. For example, the larger the distance, the closer the inconsistency confidence score is to 1, indicating a higher probability of false opening or false closing.

[0030] The pipeline diagnostic head is designed based on a graph neural network decoder. This design embeds the physical pipeline structure as prior knowledge into the model, enabling fault localization to specific nodes. First, the fire protection pipeline network is abstracted as a graph. Each node represents the physical location of a pipeline point (e.g., pipe junction, fire hydrant, valve). Node features include static attributes (design pipe diameter, elevation) and dynamic data (historical average pressure). Edges represent pipe connections, and edge features include pipe scalar values ​​such as pipe length, roughness, and material. A 2-4 layer graph convolutional network is deployed, with the number of layers related to the pipeline topology diameter. This ensures that information can be transmitted to affected nodes through the connecting edges. Insufficient layers will prevent remote nodes from detecting faults; too many layers will cause excessive smoothing of node features and increase computational load. For example, a large pipeline network may require 3 layers. After graph convolution, each node obtains a feature vector containing global topological information. At this point, a fully connected layer is applied to each node, outputting a scalar value. This scalar value, processed by the sigmoid function, is transformed into a fault confidence score (0-1) for that node. Based on this fault confidence score, the probability of a blockage or leak occurring at that point is independently determined. Finally, the output is a list of probabilities, with a length equal to the number of nodes, where each value corresponds to a node's fault confidence score. This design achieves parallel node-level evaluation, outputting the state of all nodes in a single forward propagation, resulting in extremely high efficiency. The end-to-end multi-task learning framework is trained centrally on a server or high-performance computing platform. It requires a large-scale labeled historical dataset covering the outputs of the three task decoders to form sample pairs of {multi-source input data, multi-task labels}. It adopts progressive training, first letting the model learn simple tasks (such as pump group classification) and then gradually adding more complex tasks (such as pipeline network location) to improve training stability. It uses a weighted multi-task loss function, and the weights can be adjusted according to the importance of the task and the amount of data.

[0031] The trained model is deployed on an edge computing industrial server in the fire control room and outputs all evaluation results at once. The multi-task learning framework generates a structured diagnostic result that includes pump group health status assessment results, valve consistency identification results, and a list of pipeline node fault confidence by aggregating the outputs of the three task decoders.

[0032] The structured diagnostic results will be displayed as output on the interactive interface for operators to perform checks; it serves as an operation and maintenance management output for the system. Some data from the structured diagnostic results will be used subsequently, but this will not affect the characteristic of the structured diagnostic results as monitoring outputs.

[0033] To address the challenges of distinguishing between normal equipment aging and sudden malfunctions, and the fact that environmental fluctuations (such as seasonal changes and water source variations) can easily mask or misrepresent the true health status of equipment, leading to false alarms or missed alarms, this design establishes an initial health baseline and aging trajectory function to map a unique health aging curve for each piece of equipment. Furthermore, by using a condition influence lookup table, the impact of external factors such as temperature and water source on monitoring indicators is quantitatively removed, generating corrected health indicators. These indicators eliminate reversible environmental interference, truly reflecting the irreversible performance degradation of the equipment. Simultaneously, the adaptive alarm threshold dynamically adjusts with equipment aging, avoiding false alarms caused by normal equipment aging while maintaining high sensitivity to accelerated degradation. This achieves truly personalized and precise health management. The specific design is as follows: The method for dynamically generating adaptive alarm thresholds includes: Quantitative features of the health status of each key device are extracted from the structured diagnostic results and used as core health indicators. In the initial stage of fire protection facility operation, core health indicators are collected under multiple operating conditions (referring to the state under different external conditions and internal operating modes; to ensure the universality of the initial health baseline, data on different seasons, water sources, loads, and times need to be systematically collected when the equipment is brand new). Initial health baseline vectors are established for each key piece of equipment (including fire pumps, motors, pipelines, and valve actuators). The aim is to define a health standard for each type of equipment. The method is as follows: In the initial stage of operation (e.g., the first 3 months), the quantitative outputs from the structured diagnostic results are horizontally spliced ​​to obtain the initial health baseline vector. During equipment operation, core health indicators are recorded in chronological order to form a historical sequence of health indicators. Based on historical health indicator sequences, regression analysis is used to fit the aging trend of equipment health status over operating time, resulting in an equipment aging trajectory function. Specifically, based on the physical laws of equipment wear, exponential decay functions or linear functions are often chosen. For example, an aging trajectory function HI(t) = a × exp(b × t) + c is fitted, where t is the operating time, HI(t) is the predicted health indicator, and a, b, and c are the fitting parameters. The equipment aging trajectory function describes the natural decline trend of equipment health under conditions without sudden failures. Reversible operating condition fluctuations are extracted from real-time health indicators, and corrected health indicators reflecting the inherent performance degradation of the equipment are output. Based on the equipment aging trajectory function, the corresponding adaptive alarm threshold for each core health indicator is dynamically calculated and updated. Specifically, for the current time t_now, the predicted health index HI(t_now) is calculated using the aging trajectory function. Based on the fluctuation of historical data (e.g., standard deviation σ_hist), a dynamic alarm threshold is set. For example, alarm threshold = HI(t_now) + k × σ_hist (k is the sensitivity coefficient, such as 3). As new HI data points are added to the historical sequence, the aging trajectory function is periodically refitted and σ_hist is updated, thereby achieving adaptive updating of the threshold.

[0034] The method for separating reversible operating condition fluctuations from real-time health indicators and outputting corrected health indicators that reflect the inherent performance degradation of the equipment includes: Establish a condition impact lookup table with season, water source, and ambient temperature and humidity as inputs, and typical deviations of health indicators as outputs. Specifically, the condition impact lookup table is an empirical database used to quantify, store, and query the impact of operating conditions on equipment health indicators. Example: Combination number Ambient temperature Water source type Typical vibration values ​​(mm / s) Offset relative to the reference (mm / s) (Benchmark) Room temperature (10-30°C) Municipal water supply 2.1 0.0 1 High temperature (>30°C) Municipal water supply 2.3 +0.2 2 Low temperature (<10°C) Municipal water supply 2.5 +0.4 3 Room temperature (10-30°C) Water supply from reservoirs 2.4 +0.3 4 High temperature (>30°C) Water supply from reservoirs 2.6 +0.5 5 Low temperature (<10°C) Water supply from reservoirs 2.8 +0.7 The offset of the current operating condition is obtained by looking up the operating condition influence lookup table, and the real-time core health indicators are corrected (specifically, if the offset increases, the original data is subtracted, and if the offset decreases, the original data is added). This is recorded as the corrected health indicator after removing reversible operating condition fluctuations.

[0035] To address the issues of traditional operations and maintenance (O&M) relying on manual experience to assess risks, resulting in delayed and non-standardized decisions, it is necessary to automate and standardize the transformation of technical diagnostic conclusions into actionable O&M decisions. This is achieved by combining corrected health indicators, aging rates, alarm thresholds, and external security levels through multi-factor weighted quantification, outputting a quantitative risk level. Based on this level, a pre-defined policy library is queried to automatically generate O&M policy instructions containing specific parameters. This transforms complex equipment status assessments into clear risk levels and explicit action instructions, shifting O&M work from reactive or fixed-cycle planning to precise predictive maintenance based on real-time risks. This significantly improves the scientific rigor and proactiveness of O&M management. The specific design is as follows: The method for conducting a quantitative risk assessment based on fire safety levels and generating operation and maintenance strategy instructions that include trial operation and maintenance parameters includes: Obtain the aging rate represented by the equipment aging trajectory function. Specifically, if the aging trajectory function is a linear function, the aging rate = slope; if it is an exponential function, the aging rate = c × b × exp(b × t). At a specific moment t = t_now, the instantaneous aging rate at the current moment can be obtained. Combined with the current fire safety level requirements, compare the corrected health index with the adaptive alarm threshold and evaluate the aging rate to quantify the failure risk of individual equipment and the overall fire protection system, outputting a quantified risk level. Specifically, calculate the ratio S_h of the corrected health index (HI_real) to the adaptive alarm threshold (Threshold). The larger S_h is than 1, the higher the risk. Calculate the ratio S_t of the current aging rate (V_now) to the baseline aging rate (such as the initial slope V_base of the fitted function). The larger S_t is than 1, the more significant the accelerated deterioration.

[0036] The coefficient is set according to the current fire safety level requirements of the building or area. For example: Level 1 risk areas (such as chemical plants and data centers): weight W_g = 1.5; Level 2 risk areas (such as shopping malls and warehouses): weight W_g = 1.2; Level 3 risk area (ordinary office building): Weight W_g = 1.0; Calculate the overall risk value R, for example: R = W_g × sqrt(S_h × S_t); Risk levels are quantified based on the comprehensive risk value R. For example: Low risk (R<1.2): Normal condition or slight aging, conduct scheduled inspections.

[0037] Medium risk (1.2≤R<1.8): Signs of deterioration have appeared, requiring enhanced monitoring or preparation for maintenance.

[0038] High Risk (R≥1.8): A clear risk of failure or accelerated deterioration exists, requiring immediate intervention. Based on the quantified risk level, the pre-stored operation and maintenance (O&M) strategy library is queried to generate corresponding O&M strategy instructions. These instructions must include at least the trial operation configuration parameters for fire pumps and maintenance recommendations for equipment with identified latent faults. Specifically, the O&M strategy library is a structured rule base used to establish the mapping relationship between quantified risk levels, equipment types, and specific O&M actions. Example: Strategy ID Applicable equipment types Risk level Operation and maintenance actions Trial run configuration parameters Maintenance Recommended Parameters P-101 fire pump Low risk Regular trial operation Frequency: Weekly Duration: 10 minutes Load: Idle none P-102 fire pump Medium risk Strengthen monitoring and trial operation Cycle: Every 3 days Duration: 15 minutes Load: Low load (bypass valve open) Vibration analysis is recommended to be scheduled within 2 weeks. P-103 fire pump High risk Check immediately and prepare to switch. Cycle: Immediately conduct a diagnostic trial run. It is recommended to shut down the machine for inspection within 24 hours; Automatically trigger standby pump switchover ready command V-201 Electric valve Medium risk Valve action test Action: Repeat the short stroke 3 times. It is recommended to check the valve stem seal during the next maintenance. N-301 Pipeline pressure nodes High risk None (monitoring only) none It is recommended to immediately investigate the leak point in the pipeline network of this area. To ensure the safe and reliable implementation of intelligent diagnostics and decision-making, and to absolutely guarantee the highest priority response capability of the fire protection system in emergencies, software policy commands are translated into automatic control commands for physical equipment (pumps, valves), enabling automatic trial operation and main / backup switching. Simultaneously, a highest-priority interruption mechanism is set up to ensure that all tests are immediately halted upon fire alarm or manual operation, unconditionally guaranteeing fire suppression functionality. The specific design is as follows: The method for converting operation and maintenance policy instructions into control instructions includes: During non-fire periods, the fire pumps are automatically put into trial operation according to the trial operation configuration parameters, and active excitation and response tests are triggered simultaneously during the trial operation. When the calibrated health indicators fall below their corresponding adaptive alarm thresholds, or the quantified risk level reaches the preset high-risk level, the main and backup equipment is automatically switched, and enhanced health checks (referring to active excitation trial operation) are performed on the switched main equipment. Specifically, this is an automated safety takeover process triggered by decisions and executed by hardware. When any condition is met, the control core immediately generates a main and backup equipment switching command, which is a set of predefined, standard industrial control signals. This is achieved by controlling the electrical circuits within the fire pump control cabinet, as shown in the example below: Disconnect the main pump control circuit: Send a "open" or "stop" command to the main pump's motor control circuit (usually a contactor or soft starter).

[0039] Close the standby pump control circuit: After confirming that the main pump has stopped, send a "close" or "start" command to the standby pump control circuit.

[0040] Switching pipeline valves (if required by system design): Automatically switch electric valves to guide water flow from the original main pump outlet pipeline to the standby pump outlet pipeline.

[0041] After the switch is completed, the operating status (current, pressure) of each device is fed back to the control core to confirm the successful switch.

[0042] Upon receiving a fire alarm signal or a manual forced pump start command (from the physical emergency pump start button on the fire pump control cabinet or fire control room console), all active excitation tests shall be immediately stopped.

[0043] Given the limited edge computing resources and varying monitoring needs under different states, resource allocation needs to be optimized. Simultaneously, to avoid information overload, different levels of risk must be communicated to the relevant responsible parties in the most effective way. A dual-threshold trigger mechanism is used: a low sampling rate sleep mode is employed to conserve energy when the state is stable, while a high sampling rate is immediately switched to wake-up and deep detection upon the appearance of abnormal signs. By defining tiered alarm push rules, risks are matched with response levels, as specifically designed below: The method for adaptively adjusting the data acquisition strategy based on equipment stability includes: when the variation amplitude of the monitored values ​​corresponding to the core health indicators for multiple consecutive periods is less than a first preset threshold, the fire protection facility is determined to be in stable condition, and data acquisition is performed at a low sampling rate (e.g., reducing the sampling rate from 25.6 kHz to 1 kHz). Specifically, the first preset threshold is used to determine whether the equipment is in a long-term stable operating state, and can be set according to the standard deviation Δstable of the core health indicator's variation value within a continuous period. For example, the first preset threshold is usually set as k1 × σ_stable, where k1 is a small coefficient (e.g., 1.5 or 2). If the daily fluctuation standard deviation Δstable of a water pump bearing vibration health indicator during the stable period is 0.05, and k1 is set to 2, then the first preset threshold is 0.1. If the daily vibration value variation is within ±0.1 for a consecutive week, then its condition is determined to be stable. When the amplitude or current value of the vibration signal collected in real time exceeds the second preset threshold (the setting process is similar to the first preset threshold), an abnormal sign is determined to have occurred. The sampling rate is then automatically increased (e.g., the sampling rate is increased to 25.6 kHz), and an active excitation test is triggered.

[0044] The method for issuing tiered alerts based on quantitative risk levels includes: Low-level risks are pushed to maintenance work orders, medium-level risks are notified to on-duty personnel via mobile messages, and high-level risks trigger fire alarm signals and are uploaded to the superior management platform.

[0045] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

[0046] It should be noted that all formulas in this manual are calculated by removing dimensions and taking their numerical values. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters and thresholds in the formulas are set by those skilled in the art according to the actual situation.

[0047] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A smart operation and maintenance management and control system for the entire life cycle of fire protection facilities, characterized in that, include: The data acquisition and active response testing module is used to collect the operating status of fire pumps, motors, pipelines and valve actuators through multi-source sensing to obtain real-time field status data; and in non-fire conditions, if the main fire pipeline meets the pressure conditions, it performs active excitation and response testing to obtain dynamic response data. The health diagnosis and latent fault identification module adopts an end-to-end multi-task learning framework to process field status data and dynamic response data, and outputs structured diagnostic results including the health status of pump sets, valves and pipelines. The full lifecycle management module is used to establish the initial health baseline of each key device, construct a historical sequence of health indicators based on structured diagnostic results, fit the device aging trajectory function, and dynamically generate adaptive alarm thresholds. The risk assessment and strategy generation module is used to generate a quantitative risk level by combining the fire safety level, and to generate operation and maintenance strategy instructions that include trial operation and maintenance parameters. The intelligent linkage control execution module is used to convert operation and maintenance strategy instructions into control instructions, automatically execute equipment master / slave switching and strong health detection, and prioritize the protection of fire protection functions when a fire alarm or manual forced command is received. The adaptive optimization and alarm module is used to adaptively adjust the data acquisition strategy according to the stability of the equipment status and execute graded alarms according to the quantified risk level.

2. The intelligent operation and maintenance management and control system for the entire life cycle of fire protection facilities according to claim 1, characterized in that, The method for multi-source sensing acquisition of the operating status of fire pumps, motors, pipelines, and valve actuators includes: Multiple acceleration sensors are installed on the bearing housing and pump body of the fire pump to collect vibration signals of the bearing and pump body. The three-phase current, voltage, and active power waveforms of the fire pump motor are collected using electrical parameter sensors. Pressure sensors and flow meters are installed at key nodes of the fire protection pipeline network to collect pressure and flow data of the network. The limit switch status, encoder position signal, and execution torque curve of the electric valve actuator are collected by position and torque sensors.

3. The intelligent operation and maintenance management and control system for the entire life cycle of fire protection facilities according to claim 2, characterized in that, The method for performing active stimulus and response testing includes: Determine if the pressure in the main fire hydrant pipeline is not lower than the preset lower limit; if so, then... Controlled flow or pressure disturbances are injected into the fire pipeline by briefly adjusting the speed of the fire pump or briefly controlling the opening of the bypass drain valve. Perform short-stroke reciprocating opening and closing actions at a preset angle or apply short-stroke torque loading to key valves; During the active incentive execution, multi-source sensor data is collected to obtain dynamic response data; When the pressure in the main fire pipeline is detected to be lower than the preset lower limit, the active excitation and response test will be stopped. Furthermore, real-time on-site status data refers to data acquired through multi-source sensing during the full lifecycle management of fire protection facilities.

4. The intelligent operation and maintenance management and control system for the entire life cycle of fire protection facilities according to claim 3, characterized in that, The end-to-end multi-task learning framework includes a shared input layer, one shared encoder, and three dedicated task decoders. The shared input layer receives field status data and dynamic response data. The three dedicated task decoders include a pump health head, a valve identification head, and a pipeline diagnostic head. The pump health head is designed based on a multilayer perceptron, including an input layer, hidden layers, and an output layer. The input layer receives the output of the shared encoder. The hidden layers consist of 2 to 4 fully connected layers, with the number of neurons decreasing progressively in each layer. The hidden layers are activated using the ReLU function. The output layer outputs the bearing wear level and the degree of pump corrosion. If the bearing wear level is predefined as N, then the number of neurons in the output layer is N+1. The Softmax function is used to activate the bearing wear level task, outputting the probability of each level; the Sigmoid function is used to activate the pump corrosion degree task, outputting a value between 0 and 1. The valve recognition head is designed based on a contrastive learning structure, including a two-branch Siamese network. Each branch is a 3-4 layer multilayer perceptron, and the two branches share weights. The input of branch one is the valve's stroke feedback signal features; the input of branch two is the pressure-flow response features of the pipeline network within the same time period. The two sub-networks output vectors of the same dimension. Then, the distance between the two branch vectors is calculated. This distance is converted into an inconsistency confidence score with a value range of 0-1 through a fully connected layer and a sigmoid function. The pipeline diagnostic head is designed based on a graph neural network decoder. First, the fire pipeline network is abstracted as a graph, where each node represents the physical location of a pipeline point. Node features include static attributes and dynamic data, and edges represent pipeline connections. Edge features include pipeline scalar values. A 2-4 layer graph convolutional network is deployed, with the number of layers related to the diameter of the pipeline topology. After graph convolution, each node obtains a feature vector containing global topology information. At this point, a fully connected layer is connected to each node, outputting a scalar value for each node. After processing by the sigmoid function, this value is transformed into the fault confidence of that node. The output is a probability list, with the list length equal to the number of nodes. Each value in the list corresponds to the fault confidence of a node. The multi-task learning framework generates a structured diagnostic result by aggregating the outputs of three task decoders. This result includes pump group health status assessment results, valve consistency identification results, and a list of pipeline node fault confidence.

5. The intelligent operation and maintenance management and control system for the entire life cycle of fire protection facilities according to claim 4, characterized in that, The method for dynamically generating adaptive alarm thresholds includes: Quantitative features of the health status of each key device are extracted from the structured diagnostic results and used as core health indicators. In the initial stage of fire protection facility operation, core health indicators under multiple working conditions are collected to establish the initial health baseline vector of each key equipment. During equipment operation, core health indicators are recorded in chronological order to form a historical sequence of health indicators. Based on the historical sequence of health indicators, regression analysis is used to fit the aging trend of equipment health status with operating time, and the equipment aging trajectory function is obtained. Reversible operating condition fluctuations are extracted from real-time health indicators, and corrected health indicators that reflect the inherent performance degradation of the equipment are output. Based on the equipment aging trajectory function, the corresponding adaptive alarm threshold is dynamically calculated and updated for each core health indicator.

6. The intelligent operation and maintenance management and control system for the entire life cycle of fire protection facilities according to claim 5, characterized in that, The method for separating reversible operating condition fluctuations from real-time health indicators and outputting corrected health indicators that reflect the inherent performance degradation of the equipment includes: Establish a lookup table for the impact of operating conditions, with season, water source, and ambient temperature and humidity as inputs and typical deviations of health indicators as outputs; The offset of the current operating condition is obtained by looking up the operating condition influence lookup table, and the real-time core health indicators are corrected. This is recorded as the corrected health indicator after removing reversible operating condition fluctuations.

7. The intelligent operation and maintenance management and control system for the entire life cycle of fire protection facilities according to claim 6, characterized in that, The method for generating a quantitative risk level by combining fire safety level and generating operation and maintenance strategy instructions that include trial operation and maintenance parameters includes: Obtain the aging rate represented by the device aging trajectory function; Based on the current fire safety level requirements, the corrected health indicators are compared with the adaptive alarm threshold, and the aging rate is evaluated, thereby quantifying the failure risk of individual equipment and the overall fire protection system and outputting the quantitative risk level. Based on the quantified risk level, the pre-stored operation and maintenance strategy library is queried to generate corresponding operation and maintenance strategy instructions. The operation and maintenance strategy instructions shall at least include the trial operation configuration parameters for fire pumps and the maintenance suggestion parameters for equipment with identified hidden faults.

8. The intelligent operation and maintenance management and control system for the entire life cycle of fire protection facilities according to claim 7, characterized in that, The method for converting operation and maintenance policy instructions into control instructions includes: During non-fire periods, the fire pumps are automatically put into trial operation according to the trial operation configuration parameters, and active excitation and response tests are triggered simultaneously during the trial operation. When the corrected health indicators are lower than their corresponding adaptive alarm thresholds, or the quantified risk level reaches the preset high-risk level, the main and backup equipment are automatically switched, and enhanced health detection is performed on the switched main equipment. Upon receiving a fire alarm signal or a manual forced pump start command, all active excitation tests shall be immediately stopped.

9. The intelligent operation and maintenance management and control system for the entire life cycle of fire protection facilities according to claim 8, characterized in that, The method for adaptively adjusting the data acquisition strategy based on the stability of the equipment status includes: when the change range of the monitoring value corresponding to the core health indicator for multiple consecutive cycles is less than the first preset threshold, the fire protection facility is determined to be stable, and data acquisition is carried out at a low sampling rate. When the amplitude or current value of the vibration signal collected in real time exceeds the second preset threshold, an abnormal sign is determined, the sampling rate is automatically increased, and an active excitation test is triggered.

10. The intelligent operation and maintenance management and control system for the entire life cycle of fire protection facilities according to claim 9, characterized in that, The method for issuing tiered alerts based on quantitative risk levels includes: Low-level risks are pushed to maintenance work orders, medium-level risks are notified to on-duty personnel via mobile messages, and high-level risks trigger fire alarm signals and are uploaded to the superior management platform.

Images