A fault prediction method and device for a fresh air air conditioning system

Abstract

The application discloses a kind of new air conditioning system's fault prediction method and device. The method comprises the following steps: the multi-source sensor is used to collect the multi-source perception parameter of new air conditioning system;Data compression forms standardized data packet;Data packet is decompressed and key features are extracted, to combine sensor topology graph, output failure probability matrix and space-time feature vector based on space-time graph convolution network, output the influence coefficient of external environment parameter on system performance based on environmental coupling model;Failure probability matrix is combined with historical environmental data to input bayesian decision tree model together, to output the confidence of each type of fault;And, space-time feature vector is input life prediction model, to output residual life probability distribution;Based on the confidence of each type of fault and residual life probability distribution, visual interface is displayed, and alarm threshold is dynamically adjusted.

Description

TECHNICAL FIELD

[0001] The application relates to a fault prediction method for a fresh air conditioning system, and also relates to a corresponding fault prediction device, and belongs to the technical field of air conditioning. BACKGROUND

[0002] The current fresh air conditioning system generally adopts a traditional maintenance mode of "periodic manual inspection", which lacks real-time and continuous performance monitoring means. Since the equipment state cannot be grasped at any time, the operation and maintenance personnel can only check on site according to a fixed cycle, so that potential problems are difficult to discover in time.

[0003] In the above mode, the system alarm mainly relies on a "single sensor threshold alarm" mechanism: a fixed threshold is set on the vibration or current sensor, and an alarm is triggered once the real-time value exceeds the threshold. The advantage of this scheme is low hardware cost and simple implementation, but the disadvantage is also significant - environmental interference (such as pressure difference fluctuation caused by change of air valve opening) is easy to cause misjudgment, and the false alarm rate is high; at the same time, it can only issue an alarm when the value has exceeded the standard, and cannot predict the occurrence of failure in advance.

[0004] In addition, the current "periodic manual inspection" can only rely on the experience of operation and maintenance personnel to conduct visual or aural inspection of the equipment, although obvious visible damage can be found, early and hidden performance degradation cannot be identified, and it is also difficult to quantitatively evaluate the health state of the equipment. On the other hand, some fresh air conditioning systems attempt to introduce a "simple data analysis" method, that is, based on historical operation data for statistics to determine whether there is an anomaly. However, this kind of method often only analyzes a single data source in isolation, lacks the ability to cooperatively process multiple source signals such as vibration, current, and environment, resulting in one-sided analysis results that are difficult to fully and accurately reflect the true health level of the equipment.

[0005] Under the superimposed effect of the above operation difficulties, "after-the-fact repair and shutdown replacement" has become the norm - bearings are not discovered until they are worn out, filters are not alarmed until they are clogged and the air volume is reduced, and motors are not shut down until they are overheated and burned due to dust accumulation. Each sudden failure not only brings high emergency repair costs, but also causes the fresh air conditioning system to be shut down for a long time, directly affecting indoor air quality and user health. Therefore, it is necessary to change from "passive repair" to "active prevention" by using real-time data-driven fault prediction technology to identify and quantify the remaining life when the equipment performance just shows signs of degradation, which has become an inevitable choice to reduce operation and maintenance costs, shorten downtime, and ensure the continuous and healthy operation of the system. SUMMARY

[0006] The primary technical problem to be solved by the application is to provide a fault prediction method for a fresh air conditioning system.

[0007] Another technical problem to be solved by the present application is to provide a fault prediction device for a fresh air conditioning system.

[0008] To achieve the above technical purposes, the present application adopts the following technical solutions:

[0009] According to a first aspect of an embodiment of the present application, a fault prediction method for a fresh air conditioning system is provided, comprising the following steps:

[0010] Collecting vibration signals, current signals and external environmental parameters of the fresh air conditioning system through multi-source sensors;

[0011] Performing frequency domain sparsification on the vibration signals to extract vibration sparse features, performing harmonic analysis on the current signals to extract current compression features, and performing vector normalization on the external environmental parameters to extract environmental state vectors;

[0012] Packaging, encrypting and state identifying the vibration sparse features, current compression features and environmental state vectors to form a standardized data packet;

[0013] Decrypting the standardized data packet and extracting key features to combine a sensor topology graph, capture the working mode of vibration signals and current signals in spatial distribution and time sequence based on a spatio-temporal graph convolution network, thereby outputting a fault probability matrix and a spatio-temporal feature vector; and analyzing the influence of external environmental parameters on system performance based on an environmental coupling model to output an influence coefficient;

[0014] Inputting the fault probability matrix into a Bayesian decision tree model and combining historical environmental data to perform risk assessment to output the confidence of each type of fault; and inputting the spatio-temporal feature vector into a life prediction model to output a residual life probability distribution;

[0015] Based on the confidence of each type of fault and the residual life probability distribution, the system running state, fault trend and maintenance scheme are displayed through a three-dimensional graphical interface; and based on the influence coefficient, the alarm threshold of the fresh air conditioning system is dynamically adjusted.

[0016] Preferably, the fault probability matrix and the spatio-temporal feature vector are output based on the spatio-temporal graph convolution network, specifically including:

[0017] The vibration sparse features, current compression features and sensor topology graph are jointly input into the spatio-temporal graph convolution network; wherein the sensor topology graph is used to describe the physical position relationship, signal transmission path and time delay characteristics of each sensor on the system;

[0018] The correlation between bearing vibration and impeller vibration is captured through spatial convolution, and the harmonic amplitude rising trend is analyzed through time convolution, to output bearing wear probability, blade dust accumulation probability, filter blockage probability and other fault probabilities; and the sensor topology graph is provided as a physical correlation basis for the space-time graph convolution network, to output a space-time feature vector;

[0019] The bearing wear probability, blade dust accumulation probability, filter blockage probability and other fault probabilities together form a fault probability matrix.

[0020] Preferably, the confidence levels of the various types of faults include bearing wear probability confidence level, blade dust accumulation probability confidence level, filter blockage probability confidence level and other fault probability confidence level.

[0021] Preferably, the life prediction model is constructed by the following steps:

[0022] Collecting and preprocessing multi-source data; wherein the multi-source data at least includes vibration signals, current waveforms and environmental parameters;

[0023] Performing spatial graph convolution and time convolution on the preprocessed data to output a space-time feature vector;

[0024] Inputting the space-time feature vector into an initial physical information neural network model to output residual life distribution parameters; wherein the residual life distribution parameters at least include control attenuation degree, Weibull shape parameter and baseline life;

[0025] Based on historical data, the physical parameters are inversely solved and parameter labels are generated to form a plurality of training samples, thereby collectively constituting a training set;

[0026] Based on the training set, the initial physical information neural network model is trained to form the life prediction model.

[0027] Preferably, the alarm threshold of the fresh air conditioning system is dynamically adjusted by the following formula:

[0028] Wherein, represents the alarm threshold at time t; T base represents the initial alarm threshold; r represents the maximum adjustment amplitude; represents a smooth transition; represents the influence coefficient at time t; k represents the change rate sensitivity coefficient of the environment; s represents the normalization factor, which is a fixed parameter; represents the change rate of the environment.

[0029] Preferably, the fault prediction method further comprises:

[0030] generate system maintenance instructions based on the residual life probability distribution and the dynamically adjusted alarm threshold;

[0031] The system maintenance instructions at least include no action, cleaning, or replacement.

[0032] According to a second aspect of the embodiments of the present application, a fault prediction device for a fresh air conditioning system is provided, comprising:

[0033] A perception layer is configured to collect vibration signals, current signals, and external environment parameters of the fresh air conditioning system through multi-source sensors, and perform frequency domain sparsification on the vibration signals to extract vibration sparse features, perform harmonic analysis on the current signals to extract current compression features, and perform vector normalization on the external environment parameters to extract an environment state vector.

[0034] A transmission layer is connected to the perception layer and configured to encapsulate, encrypt, and state-identify the vibration sparse features, the current compression features, and the environment state vector to form a standardized data packet.

[0035] An analysis layer is connected to the transmission layer and configured to decrypt the standardized data packet to extract key features, and perform data analysis in combination with a sensor topology graph to output a fault probability matrix, a spatio-temporal feature vector, and an influence coefficient of external environment parameters on system performance.

[0036] A decision layer is connected to the analysis layer and configured to make a decision based on the fault probability matrix, the spatio-temporal feature vector, and the influence coefficient to output confidence levels of various faults and a residual life probability distribution.

[0037] An application layer is connected to the decision layer and configured to display system running states, fault trends, and maintenance schemes through a three-dimensional graphical interface based on the confidence levels of various faults and the residual life probability distribution, and the application layer is also connected to the analysis layer and configured to dynamically adjust an alarm threshold of the fresh air conditioning system based on the influence coefficient.

[0038] Preferably, the analysis layer comprises:

[0039] A fault analysis module is preconfigured with a spatio-temporal graph convolution network to receive the key features and the sensor topology graph, capture working modes of the vibration signals and the current signals in spatial distribution and time series, and output the fault probability matrix.

[0040] A performance influence module is preconfigured with an environment coupling model to receive the key features and analyze influences of external environment parameters on system performance, and output the influence coefficient.

[0041] Preferably, the decision layer comprises:

[0042] a risk assessment module, which is provided with a Bayesian decision tree model, receives the fault probability matrix, and performs risk assessment in combination with historical environment data, thereby outputting confidence levels of various faults;

[0043] a life prediction module, which is provided with a life prediction model, receives the space-time feature vector, and outputs a residual life probability distribution through the life prediction model.

[0044] Preferably, the application layer comprises:

[0045] a visualization module, which is connected to the decision layer, and displays system operation states, fault trends and maintenance schemes through a three-dimensional graphical interface based on the confidence levels of various faults and the residual life probability distribution.

[0046] a threshold adjustment module, which is connected to the analysis layer, and dynamically adjusts the alarm threshold of the fresh air conditioning system based on the influence coefficient.

[0047] Compared with the prior art, the present application has the following technical effects:

[0048] (1) Through data acquisition, feature extraction, state analysis, risk prediction and control feedback to form a closed-loop management, the performance degradation of the fresh air conditioning system can be identified early, and the residual life of the fresh air conditioning system can be predicted, thereby providing a guidance direction for the maintenance of the fresh air conditioning system, and avoiding the blindness of system maintenance.

[0049] (2) Through the multi-source physical signals collected by the multi-source sensors during the operation of the fresh air conditioning system, fault analysis can be performed from multiple dimensions, and the collaborative prediction of multiple types of faults (such as bearing wear, blade dust accumulation, filter blockage, etc.) can be realized.

[0050] (3) Through the environmental coupling model to analyze the influence of external environmental parameters on the system performance, and then dynamically adjust the alarm threshold of the fresh air conditioning system according to the influence coefficient, thereby avoiding the problems of false alarm or missed alarm of the system caused by environmental interference. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 a flowchart of the fault prediction method for the fresh air conditioning system provided by the first embodiment of the present application;

[0052] Figure 2 a structural diagram of the fault prediction device for the fresh air conditioning system provided by the second embodiment of the present application;

[0053] Figure 3 a structural diagram of the perception layer in the second embodiment of the present application;

[0054] Figure 4As shown in the structural diagram of the analysis layer in the second embodiment of the present application,

[0055] Figure 5 As shown in the structural diagram of the decision layer in the second embodiment of the present application,

[0056] Figure 6 As shown in the structural diagram of the application layer in the second embodiment of the present application,

[0057] Figure 7 As shown in the structural diagram of the fault prediction device for the fresh air air conditioning system provided in the third embodiment of the present application. DETAILED DESCRIPTION

[0058] The technical content of the present application will be described in detail below in combination with the drawings and specific embodiments.

[0059] The embodiments of the present application provide a fault prediction method and system for a fresh air air conditioning system. The entire system is composed of five levels of sensing layer, transmission layer, analysis layer, decision layer and application layer (as shown in Figure 2 ), to jointly realize the fault prediction method. Through data acquisition, feature extraction, state analysis, risk prediction and control feedback, a closed-loop management is formed between each level, for performance detection and fault prediction of the fresh air air conditioning system, so as to realize intelligent monitoring and operation and maintenance support for the equipment running state.

[0060] First embodiment

[0061] As shown in the structural diagram of the analysis layer in the second embodiment of the present application, Figure 1 The first embodiment of the present application provides a fault prediction method for a fresh air air conditioning system, which specifically includes the following steps:

[0062] S10: Obtain multi-source sensing parameters.

[0063] In this embodiment, multi-source physical signals of the fresh air air conditioning system in the running process are collected by multi-source sensors, and preliminary processing is performed to obtain multi-source sensing parameters.

[0064] Specifically, the multi-source sensing parameters at least include:

[0065] (1) The vibration signal of the fresh air air conditioning system is collected by the MEMS piezoelectric array, and the frequency domain sparsification is performed on the vibration signal to extract the vibration sparse feature. The feature can be used to identify the early abnormal vibration feature of the key components (such as bearings and gears) of the equipment.

[0066] (2) The current signal of the fresh air air conditioning system is collected by the intelligent current sensor, and the harmonic analysis is performed on the current signal to extract the current compression feature, which is used to detect the motor running state, load change and electrical fault precursor.

[0067] (3) Through the environmental sensor, real-time collection of environmental parameters, including PM2.5 concentration, TVOC (total volatile organic compounds), temperature, humidity, etc., and vector normalization according to external environmental parameters to extract environmental state vectors, thereby providing basic data for subsequent environmental coupling analysis.

[0068] It can be understood that the vibration sparsity feature, current compression feature and environmental state vector together constitute the multi-source perception parameter. In other embodiments, other types of feature parameters can also be added as needed.

[0069] S20: Data compression forms a standardized data packet and is uploaded to the cloud or local server.

[0070] After obtaining the multi-source perception parameter through the above step S10, encryption, compression and remote transmission are performed through the edge gateway. The LoRa WAN / 5G edge gateway supports low-power wide-area network (LoRa WAN) and high-speed 5G dual-mode communication, has a data compression capability of more than 80%, and ensures efficient and stable data upload to the cloud or local server under limited bandwidth conditions.

[0071] S30: Data packet decompression and key feature extraction for fault analysis and performance impact analysis.

[0072] After uploading the data packet to the cloud or local server through step S20, the data packet is downloaded from the cloud or local server for decompression to extract key features. It can be understood that the key features correspond to the above multi-source perception parameters.

[0073] Specifically, the data analysis process includes the following two parts:

[0074] S31: Fault analysis.

[0075] After extracting the key features, the sensor topology graph needs to be combined to capture the spatial distribution and time sequence of the vibration signal and current signal based on the spatio-temporal graph convolution network, thereby outputting a fault probability matrix and a spatio-temporal feature vector. The sensor topology graph is used to describe the physical location relationship, signal transmission path and time delay characteristics of each sensor on the system.

[0076] Specifically, the vibration sparsity feature, current compression feature and sensor topology graph are jointly input into the spatio-temporal graph convolution network.

[0077] Then, the correlation between bearing vibration and impeller vibration is captured through spatial convolution, and the harmonic amplitude rising trend is analyzed through time convolution, to output bearing wear probability, blade dust accumulation probability, filter blockage probability and other failure probabilities. Wherein, the bearing wear probability, blade dust accumulation probability, filter blockage probability and other failure probabilities together form a failure probability matrix.

[0078] Finally, the sensor topology graph is used as a physical basis for the spatio-temporal graph convolution network, so as to output a spatio-temporal feature vector.

[0079] In this embodiment, the spatio-temporal feature vector (128-dimensional floating point array) is a fusion feature representation extracted from multi-modal data by a spatio-temporal graph convolution network (STGCN), which includes the following core dimensions:

[0080]

[0081] The life prediction using the spatio-temporal feature vector is realized by a physical information neural network (PINN) and is completed in the following four steps:

[0082] Step 1: Feature mapping to physical parameters. Input: spatio-temporal feature vector, output: performance decay coefficient (h -1 ) α, Weibull distribution shape parameter k, and reference characteristic life (h) λ.

[0083] Step 2: Solve the performance decay equation.

[0084] Step 3: Generate residual life distribution.

[0085] Step 4: Failure probability prediction.

[0086] For example, the input feature vector is as follows:

[0087]

[0088] Step 1: Physical parameter calculation: ;

[0089] Step 2: Vibration correction: ;

[0090] Step 3: Life distribution: ;

[0091] Step 4: Failure probability: .

[0092] S32: Performance impact analysis.

[0093] In this embodiment, an environment coupling model is constructed in advance. When the key features are received, the environment state vector is input into the environment coupling model, so as to analyze the influence of external environment parameters on system performance, and output the influence coefficient.

[0094] The training set is formed by combining the historical environment parameters with the performance parameters of the system, and the model training is performed by using the training set, so as to form the environment coupling model, and analyze the influence of external environment parameters on system performance.

[0095] It can be understood that the output result of the fault analysis performed by the above step S31 is the fault probability matrix and the spatial feature vector, and the output result of the performance influence analysis performed by the above step S32 is the influence coefficient.

[0096] S40: Risk assessment and life prediction are performed according to the results of data analysis.

[0097] Specifically, the step S40 includes the following two parts, which are risk assessment (corresponding to S41) and life prediction (corresponding to S42).

[0098] S41: Risk assessment.

[0099] In this embodiment, the fault probability matrix is input into the Bayesian decision tree model, and the historical environment data is combined to perform risk assessment, so as to output the confidence of each type of fault.

[0100] Specifically, in this embodiment, the Bayesian decision tree model composition includes:

[0101] Model architecture: [fault probability matrix] + [historical environment data] --> [Bayesian network layer] --> [decision tree classifier] --> [confidence output] --> [risk assessment].

[0102] The core components are shown in the following table:

[0103]

[0104] Risk assessment process input data: fault probability matrix O, current environment vector , historical environment database , fault loss cost table .

[0105] Risk assessment process: environment similarity matching (output similar environment set) → prior probability calculation → likelihood function modeling → Bayesian posterior calculation (output normalized posterior probability vector) → decision tree confidence classification → risk value assessment → confidence output.

[0106] The output confidence information includes: fault type, confidence, risk level, risk value and corresponding decision.

[0107] The confidence of each type of fault includes: bearing wear probability confidence, blade dust accumulation probability confidence, filter blockage probability confidence and other fault probability confidence; and the confidence value ranges from 0 to 1.

[0108] S42: Life prediction

[0109] In this embodiment, the space-time feature vector is input into the life prediction model to output the residual life probability distribution.

[0110] The life prediction model is constructed by the following steps:

[0111] Step 1: Multi-source data acquisition and preprocessing.

[0112] Data sources: vibration signal, current waveform, environmental parameter;

[0113] Preprocessing procedure: [raw data] --> [signal alignment] --> [frequency domain analysis] --> [physical normalization]

[0114] Step 2: Space-time feature vector generation.

[0115] Input: multi-dimensional feature matrix after preprocessing,

[0116] Output: space-time feature vector (example output: [0.32, -0.15, 0.08,..., 0.21] bearing coupling ↑ 32%, THD change rate ↓ 15%, high frequency energy ↑ 8%).

[0117] Step 3: Construction of physically constrained life model.

[0118] Model architecture: physical information neural network (PINN)

[0119] Input: space-time feature vector

[0120] Output: residual life distribution parameters (Weibull distribution k, λ)

[0121] Core constraint physical equation: (air volume attenuation equation)

[0122] Feature to physical parameter mapping:

[0123] α: control attenuation speed (positively correlated with vibration energy)

[0124] k: Weibull shape parameter (k < 1 indicates early failure)

[0125] : Baseline life (determined by material properties)

[0126] Step 4: Model training

[0127] Based on historical data, inverse physical parameters to generate labels, form multiple training samples, and collectively constitute a training set for model training to ultimately train the life prediction model.

[0128] S50: Visual interface display, and dynamic adjustment of alarm threshold.

[0129] It can be understood that when the confidence of various faults and the residual life probability distribution are obtained through step S40, data processing can be performed in combination with related modules, so as to display the system running state, fault trend and maintenance scheme through a three-dimensional graphical interface.

[0130] And, based on the influence coefficient output in step S30, the alarm threshold of the fresh air conditioning system is dynamically adjusted by the following formula:

[0131] Wherein, represents the alarm threshold at time t; T base represents the initial alarm threshold; r represents the maximum adjustment amplitude, the value range is 0~0.2, and the system defaults to 0.05; represents a smooth transition; represents the influence coefficient at time t; k represents the change rate sensitivity coefficient of the environment; s represents the normalization factor, which is a fixed parameter; represents the change rate of the environment.

[0132] In addition, in the embodiment, the system maintenance instructions are generated based on the residual life probability distribution and the dynamically adjusted alarm threshold. The system maintenance instructions at least include no action, cleaning or replacement, thereby guiding the system maintenance personnel to maintain.

[0133] Second embodiment

[0134] As Figure 2 shown, based on the above first embodiment, the second embodiment of the application provides a fault prediction device for a fresh air conditioning system, which includes a perception layer 1, a transmission layer 2, an analysis layer 3, a decision layer 4 and an application layer 5.

[0135] Specifically, the perception layer 1 is responsible for collecting multi-source physical signals in the device running process and performing preliminary processing. As Figure 3As shown, the perception layer 1 includes the following three types of sensor modules: MEMS piezoelectric array 11, output sparse vibration spectrum data, used to identify early abnormal vibration characteristics of key components of the device (such as bearings, gears). Intelligent current sensor 12, extract compressed harmonic data, used to detect motor operating state, load change and electrical fault precursor. Multi-source environmental sensor 13, real-time acquisition of environmental parameters, including PM2.5 concentration, TVOC (total volatile organic compounds), temperature, humidity, etc., to provide basic data for subsequent environmental coupling analysis. The above information is uniformly collected to the perception layer output, and enters the next level.

[0136] The transmission layer 2 receives data from the perception layer 1 and encrypts, compresses and remotely transmits through the edge gateway. When the standardized data packet transmission is completed, the data enters the analysis layer 3 for modeling and feature extraction.

[0137] The analysis layer 3 performs deep learning modeling and state analysis on the uploaded data. As shown, Figure 4 As shown, the analysis layer 3 includes a fault analysis module 31 and a performance impact module 32. Among them, the fault analysis module 31 is pre-set with a spatio-temporal graph convolution network, a deep neural network based on graph structure, which is used to capture the complex patterns of device vibration, current and other signals in spatial distribution and time series, output fault probability matrix, and evaluate the health status of different components. The performance impact module 32 is pre-set with an environmental coupling model, which combines the environmental data transmitted from the perception layer, dynamically adjusts the model parameters, outputs the influence coefficient, reflects the influence of environmental changes on the performance of the device, and improves the prediction accuracy. The analysis result is transmitted as input to the decision layer for further risk judgment and life prediction.

[0138] The decision layer 4 makes a comprehensive judgment according to the results of the analysis layer 3, and generates specific operation and maintenance strategies, such as Figure 5 As shown, the decision layer 4 includes a risk assessment module 41 and a life prediction module 42. Specifically, the risk assessment module 41 is pre-set with a Bayesian decision tree model, based on Bayesian inference algorithm, combined with historical data and current state, to calculate the confidence of each type of fault occurrence, to assist in formulating priority response strategy. The life prediction module 42 is pre-set with a life prediction model (physical information neural network), which introduces a neural network with physical law constraints to predict the remaining useful life of the device, and outputs in the form of probability density function, which is used to guide the arrangement of preventive maintenance plan. The final decision result is sent to the application layer 5 to perform specific operation.

[0139] The application layer 5 is the interface between the system and the user or the control system, mainly including a visualization module 51 and a threshold adjustment module 52 (such as Figure 6(As shown). Specifically, the visualization module 51 is connected to the decision layer 4 to display the system's operating status, fault trends, and maintenance plans through a three-dimensional graphical interface based on the confidence level and remaining lifetime probability distribution of various faults. The threshold adjustment module 52 is connected to the analysis layer 3 to automatically adjust the alarm threshold based on the influence coefficient and real-time operating conditions and environmental changes, avoiding false alarms and missed alarms, and improving the system's stability and adaptability.

[0140] It is understood that the functions and connections of each module unit in this embodiment are only one specific implementation of the fault prediction method in the first embodiment. In other embodiments, the functions and connections of each module unit can be adaptively adjusted as needed to achieve the fault prediction method described above.

[0141] Third Embodiment

[0142] like Figure 7 As shown, based on the above-described fault prediction method for fresh air conditioning systems, the third embodiment of the present invention further provides a fault prediction device for fresh air conditioning systems. This fault prediction device includes one or more processors and a memory. The memory is coupled to the processor and is used to store one or more programs. When the programs are executed by the processor, the processor implements the fault prediction method for fresh air conditioning systems described in the above embodiment.

[0143] The processor controls the overall operation of the fault prediction device to complete all or part of the steps of the fault prediction method for the fresh air conditioning system described above. The processor can be a central processing unit (CPU), graphics processing unit (GPU), field-programmable gate array (FPGA), application-specific integrated circuit (ASIC), digital signal processing (DSP) chip, etc. The memory stores various types of data to support the operation of the fault prediction device. This data may include, for example, instructions for any application or method operating on the fault prediction device, as well as application-related data. The memory can be implemented using any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, etc.

[0144] In one example embodiment, the fault prediction apparatus can be embodied by a computer chip or entity, or a product with certain functions, for executing the above-mentioned fault prediction method for fresh air conditioning system, and achieving the technical effects consistent with the above-mentioned method. A typical embodiment is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, an in-vehicle human-machine interaction device, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

[0145] In another example embodiment, the present application also provides a computer readable storage medium comprising program instructions, which, when executed by a processor, implement the steps of the above-mentioned fault prediction method for fresh air conditioning system in any one of the embodiments. For example, the computer readable storage medium can be the above-mentioned memory comprising program instructions, which can be executed by the processor of the fault prediction apparatus to complete the above-mentioned fault prediction method for fresh air conditioning system, and achieve the technical effects consistent with the above-mentioned method.

[0146] It should be noted that the above-mentioned embodiments are only illustrative, and the technical solutions of each embodiment can be combined, all of which are within the protection scope of the present application.

[0147] In addition, the terms "first", "second" are only for descriptive purposes, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first", "second" can explicitly or implicitly include one or more of the features. In the description of the present application, the meaning of "multiple" is two or more, unless otherwise specifically limited.

[0148] The above-mentioned fault prediction method and apparatus for fresh air conditioning system provided by the present application are described in detail. For those skilled in the art, any obvious modification made to it without departing from the essential content of the present application will constitute an infringement of the patent right of the present application, and will bear the corresponding legal responsibility.

Claims

1. A fault prediction method for fresh air conditioning systems, characterized in that... Includes the following steps: Vibration signals, current signals, and external environmental parameters of the fresh air conditioning system are collected using multi-source sensors. The vibration signal is frequency-domain sparsified to extract vibration sparsity features; Harmonic analysis is performed on the current signal to extract current compression characteristics; The external environment parameters are vector normalized to extract the environment state vector; The vibration sparsity features, current compression features, and environmental state vectors are encapsulated, encrypted, and identified to form a standardized data packet; The standardized data packets are decrypted and key features are extracted. Combined with the sensor topology map, the working patterns of vibration and current signals in spatial distribution and time series are captured based on the spatiotemporal graph convolutional network, thereby outputting the fault probability matrix and spatiotemporal feature vector. Furthermore, the impact of external environmental parameters on system performance is analyzed based on the environmental coupling model to output the impact coefficient. The method of outputting the fault probability matrix and spatiotemporal feature vector based on the spatiotemporal graph convolutional network specifically includes: inputting the vibration sparse features, current compression features, and sensor topology map into the spatiotemporal graph convolutional network; wherein, the sensor topology map is used to describe the physical positional relationship, signal transmission path, and time delay characteristics of each sensor on the system; capturing the correlation between bearing vibration and impeller vibration through spatial convolution, and analyzing the continuous upward trend of harmonic amplitude through temporal convolution to output the probability of bearing wear, blade dust accumulation, filter clogging, and other faults; and providing the spatiotemporal graph convolutional network with physical correlation evidence through the sensor topology map, thereby outputting the spatiotemporal feature vector; wherein, the probability of bearing wear, blade dust accumulation, filter clogging, and other faults together form the fault probability matrix; The failure probability matrix is ​​input into a Bayesian decision tree model and combined with historical environmental data for risk assessment to output the confidence level of various failures; and the spatiotemporal feature vector is input into a lifetime prediction model to output the remaining lifetime probability distribution. Based on the confidence level and remaining lifetime probability distribution of the various faults, the system operation status, fault trends and maintenance plans are displayed through a three-dimensional graphical interface; and, based on the influence coefficient, the alarm threshold of the fresh air conditioning system is dynamically adjusted.

2. The fault prediction method as described in claim 1, characterized in that: The confidence levels for each type of fault include: bearing wear probability confidence level, blade dust accumulation probability confidence level, filter clogging probability confidence level, and other fault probability confidence levels.

3. The fault prediction method as described in claim 1, characterized in that... The lifetime prediction model is constructed through the following steps: Collect and preprocess multi-source data; wherein the multi-source data includes at least vibration signals, current waveforms, and environmental parameters; Spatial graph convolution and temporal convolution are performed on the preprocessed data to output spatiotemporal feature vectors; The spatiotemporal feature vector is input into the initial physical information neural network model to output the remaining lifetime distribution parameters; wherein, the remaining lifetime distribution parameters include at least: the degree of control decay, the Weibull shape parameter, and the baseline lifetime; Based on historical data, physical parameters are deduced and parameter labels are generated to form multiple training samples, which together constitute the training set. The initial physical information neural network model is trained based on the training set to ultimately form the lifespan prediction model.

4. The fault prediction method as described in claim 1, characterized in that... The alarm threshold of the fresh air conditioning system is dynamically adjusted using the following formula: in, T represents the alarm threshold at time t; base Indicates the initial alarm threshold; r represents the maximum adjustment range; Indicates a smooth transition; β t represents the influence coefficient at time t; k represents the environmental change rate sensitivity coefficient; s represents the normalization factor, which is a fixed parameter. This indicates the rate of environmental change.

5. The fault prediction method as described in claim 1, characterized in that... Also includes: Based on the remaining lifetime probability distribution and the dynamically adjusted alarm threshold, system maintenance instructions are jointly generated; The system maintenance instructions include at least no action, cleaning, or replacement.

6. A fault prediction device for a fresh air conditioning system, used to execute the fault prediction method as described in any one of claims 1-5, characterized in that... include: The sensing layer is used to collect vibration signals, current signals, and external environmental parameters of the fresh air conditioning system through multi-source sensors; and to perform frequency domain sparsification on the vibration signals to extract vibration sparsity features. Harmonic analysis is performed on the current signal to extract current compression characteristics; The external environment parameters are vector normalized to extract the environment state vector; The transmission layer, connected to the sensing layer, is used to encapsulate, encrypt, and identify the vibration sparse features, current compression features, and environmental state vectors to form standardized data packets. The analysis layer, connected to the transmission layer, is used to decrypt the standardized data packets and extract key features, and to perform data analysis in conjunction with the sensor topology map, thereby outputting a fault probability matrix, spatiotemporal feature vector, and the influence coefficient of external environmental parameters on system performance. The decision layer, connected to the analysis layer, makes decisions based on the failure probability matrix, spatiotemporal feature vector, and influence coefficient, and outputs the confidence level and remaining lifetime probability distribution of various failures. The application layer, connected to the decision layer, displays the system's operating status, fault trends, and maintenance plans through a three-dimensional graphical interface based on the confidence level and remaining lifetime probability distribution of the various faults. Furthermore, the application layer is also connected to the analysis layer to dynamically adjust the alarm threshold of the fresh air conditioning system based on the influence coefficient.

7. The fault prediction device as described in claim 6, characterized in that... The analysis layer includes: The fault analysis module is pre-configured with a spatiotemporal graph convolutional network to receive the key features and sensor topology map, capture the working patterns of vibration signals and current signals in spatial distribution and time series, and thus output a fault probability matrix. The performance impact module has a preset environment coupling model to receive the key features and analyze the impact of external environmental parameters on system performance, thereby outputting the impact coefficient.

8. The fault prediction device as described in claim 7, characterized in that... The decision-making body includes: The risk assessment module is pre-set with a Bayesian decision tree model to receive the failure probability matrix and perform risk assessment in combination with historical environmental data, thereby outputting the confidence level of various failures. The lifespan prediction module has a preset lifespan prediction model to receive the spatiotemporal feature vector and output the remaining lifespan probability distribution through the lifespan prediction model.

9. The fault prediction device as described in claim 8, characterized in that... The application layer includes: The visualization module, connected to the decision-making level, displays the system's operating status, fault trends, and maintenance plans through a three-dimensional graphical interface based on the confidence level and remaining lifetime probability distribution of the various types of faults. A threshold adjustment module, connected to the analysis layer, dynamically adjusts the alarm threshold of the fresh air conditioning system based on the influence coefficient.

Images