A refrigeration system fault early warning method based on an AI model

Abstract

The application relates to an AI model-based refrigeration system fault early warning method, which comprises the following steps: S1, collecting state data, equipment design parameters and maintenance work order records; S2, removing the influence of environmental temperature factors and load factors to obtain compensated state data; S3, cross-fusing the compensated state data, the equipment design parameters and the maintenance work order records to generate comprehensive health state features; S4, establishing an FMEA failure mode identification model, an FTA fault tree construction model and an ETA event tree evolution model of the refrigeration system according to the comprehensive health state features; S5, training an AI model to obtain a refrigeration system health prediction AI model; S6, performing health state evaluation on the state data, the equipment design parameters and the maintenance work order records of the refrigeration system according to the refrigeration system health prediction AI model to generate an equipment health index; and S7, determining the corresponding fault risk priority number according to the equipment health index to generate the priority of a maintenance task and a maintenance decision.

Description

A method for early warning of refrigeration system faults based on AI models Technical Field

[0001] This invention relates to the field of refrigeration system fault early warning, specifically to a refrigeration system fault early warning method based on an AI model. Background Technology

[0002] A refrigeration system is a highly integrated cooling solution that consists of one or more cold source systems and multiple air conditioning systems. It uses the principle of liquid vaporization refrigeration to continuously and stably provide the required cooling capacity to the air conditioning system, keeping the indoor temperature at a comfortable and pleasant level.

[0003] Refrigeration systems include chillers, water pumps, cooling towers, etc., and each component contains multiple operating parameters and status parameters. To provide early warnings of potential refrigeration system failures, a fault prediction model is needed. Traditional Failure Mode and Effects Analysis (FMEA) and Fault Tree Analysis (FTA) rely heavily on human experience, lacking objectivity in risk factor scoring and difficulty in quantifying the priority of failure modes, leading to delayed maintenance decisions. For example, the existing application with publication number CN117669130A, while combining FTA and FMEA, does not incorporate dynamic data analysis, limiting its analytical efficiency. Similarly, the existing application with publication number CN118300082B relies solely on historical data to train its model, without integrating equipment maintenance experience. These existing fault prediction technologies still have limitations, lacking the ability to predict potential failures and failing to achieve health status assessment.

[0004] The purpose of this invention is to design an AI model-based fault early warning method for refrigeration systems to address the problems existing in the prior art. Summary of the Invention

[0005] To address the problems existing in the prior art, the present invention provides a refrigeration system fault early warning method based on an AI model, which can effectively solve at least one of the problems existing in the prior art.

[0006] The technical solution of this invention is:

[0007] A method for early warning of refrigeration system faults based on an AI model includes the following steps:

[0008] S1 collects multiple status data of the refrigeration system, equipment design parameters, and maintenance work order records;

[0009] S2, the state data is dynamically decoupled to remove the influence of environmental temperature and load factors, and the compensated state data is obtained.

[0010] S3, perform feature cross-fusion of the compensated status data, the equipment design parameters, and the maintenance work order records to generate comprehensive health status features;

[0011] S4. Based on the comprehensive health status characteristics, establish an FMEA failure mode identification model, an FTA fault tree construction model, and an ETA event tree evolution model for the refrigeration system. Quantify the failure risk priority of the refrigeration system through the FMEA failure mode identification model, locate the underlying cause and probability of failure of the refrigeration system through the FTA fault tree construction model, and predict the failure evolution path of the refrigeration system through the ETA event tree evolution model.

[0012] S5. The AI ​​model is trained by the comprehensive health status characteristics, the failure risk priority number and failure type output by the FMEA failure mode identification model, the underlying cause and probability of failure output by the FTA fault tree construction model, and the failure evolution path output by the ETA event tree evolution model, to obtain the refrigeration system health prediction AI model.

[0013] S6, based on the refrigeration system health prediction AI model, performs a health status assessment on the refrigeration system status data, equipment design parameters, and maintenance work order records, and generates an equipment health index;

[0014] S7. Sort the fault risk priority numbers output by the FMEA failure mode identification model, associate the range of the equipment health index with the order of the fault risk priority numbers, determine the corresponding fault risk priority numbers according to the equipment health index, and generate the priority of maintenance tasks and maintenance decisions.

[0015] Furthermore, in S1, after collecting multiple state data of the refrigeration system, the following is performed: predicting and filling in missing or abnormal values ​​in the multiple state data.

[0016] S2, the multiple state data are dynamically decoupled to remove the influence of environmental and load factors, resulting in compensated state data including:

[0017] S2.1, Establish a multiple linear regression model, and train the multiple linear regression model using ambient temperature, load rate, and the state data as inputs;

[0018] S2.2, The residual terms in the trained multiple linear regression model are used as compensation to adjust the state data to obtain compensated state data.

[0019] Further, in step S3, the compensated status data, the equipment design parameters, and the maintenance work order records are cross-fused to generate comprehensive health status features, including:

[0020] S3.1, The compensated state data and the equipment design parameters are fused to generate a deviation fusion feature with physical meaning;

[0021] S3.2, the compensated status data and the maintenance work order record are fused to generate fault fusion features, and each fault fusion feature is labeled according to the fault type in the maintenance work order record;

[0022] S3.3, assign corresponding weights to the deviation fusion feature and the fault fusion feature, and then perform weighted fusion to generate the comprehensive health status feature.

[0023] Furthermore, the fusion of the compensated status data and the maintenance work order record includes:

[0024] The compensated status data and the maintenance work order record are time-aligned using a dynamic time warping algorithm.

[0025] The fault fusion feature is obtained by fusing the compensated state data within a preset time range before and after the fault time point in the maintenance work order record, and the fault type recorded in the maintenance work order record.

[0026] Furthermore, following S3.3, it includes:

[0027] S3.4, dimensionality reduction is performed on the comprehensive health status characteristics, and the principal components with the most influence on the health assessment of the refrigeration system are selected using the PCA method.

[0028] Further, step S4 includes:

[0029] Based on the overall health status, potential failure modes of the refrigeration system are identified using the FMEA failure mode identification model. Each failure mode is assigned a severity, frequency of occurrence, and detectability. The failure risk priority number is calculated using the following formula:

[0030] ;

[0031] Based on the failure modes output by the FMEA failure mode identification model, an ETA event tree evolution model is constructed using logic gates to analyze top-level failure events, locate underlying causes of failures, and obtain the probability of occurrence of underlying causes of failures.

[0032] Based on the failure modes output by the FMEA failure mode identification model and the underlying causes and probabilities of failure output by the FTA fault tree construction model, an ETA event tree evolution model is constructed to simulate the possible paths of failure in the refrigeration system and predict the failure evolution path of the refrigeration system through the possible paths.

[0033] Further, step S5 includes:

[0034] S5.1, By using the comprehensive health status characteristics described in the history, the time-series prediction model is trained to obtain the future operating parameter prediction model, which is used to predict the future operating parameters of the refrigeration system;

[0035] S5.2, The supervised learning model is trained using the comprehensive health status features and the fault type output by the FMEA failure mode recognition model to obtain a fault probability prediction model, which is used to predict the fault type and its fault probability.

[0036] S5.3, Based on the future operating parameters predicted by the future operating parameter prediction model and the failure type probability predicted by the failure probability prediction model, establish a refrigeration system health prediction AI model.

[0037] Furthermore, establishing an AI model for predicting the health of refrigeration systems includes:

[0038] Based on the deviation between the future operating parameters predicted by the future operating parameter prediction model and the equipment design parameters, the parameter deviation health of the refrigeration system is calculated.

[0039] The equipment aging health is calculated based on the failure type probability predicted by the failure probability prediction model and the comprehensive health status characteristics.

[0040] Based on the deviation between the energy efficiency index corresponding to the future operating parameters predicted by the future operating parameter prediction model and the energy efficiency index of the equipment design, the energy efficiency deviation health of the refrigeration system is calculated.

[0041] The health status of the parameter deviation, the health status of the equipment aging, and the health status of the energy efficiency deviation are added together according to their corresponding weights to obtain the AI ​​model for predicting the health of the refrigeration system.

[0042] Further, in step S7, associating the range of the equipment health index with the order of the fault risk priority numbers includes:

[0043] The equipment health index and the fault risk priority number are normalized, and the normalized fault risk priority number is mapped to the normalized equipment health index in descending order.

[0044] Further, in step S7, the priority of generating maintenance tasks includes: prioritizing the fault risk priority number and obtaining the corresponding priority based on the fault risk priority number corresponding to the equipment health index;

[0045] Maintenance decisions for generating maintenance tasks include: preventive maintenance for fault types with low probability of failure but high impact, corrective maintenance for faults with high probability of failure and high impact, and optimization maintenance for maintenance that does not affect safety but can improve energy efficiency.

[0046] Therefore, the present invention provides the following effects and / or advantages:

[0047] This application covers all stages from data acquisition, processing, and feature fusion to fault early warning and maintenance decision generation. The solution's advantage lies in its ability to predict and assess the health of refrigeration systems through the fusion of multi-source data and the training of AI models. It also combines health status with fault risk priority to ensure a reasonable prioritization of maintenance tasks. Equipment with lower health indices and higher fault risk is prioritized for handling.

[0048] This application integrates maintenance work order records, ensuring comprehensive data monitoring of the refrigeration system. This data diversity provides a rich source of information for subsequent fault prediction and health status assessment.

[0049] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0050] It should be understood that the above summary and the following detailed description of the invention are exemplary and explanatory, and are intended to provide further explanation of the invention as claimed. Attached Figure Description

[0051] Figure 1 is a flowchart illustrating one embodiment of the present invention.

[0052] Figure 2 shows some of the failure mode codes identified by FMEA for refrigeration system efficiency degradation. Detailed Implementation

[0053] To facilitate understanding by those skilled in the art, the present invention will now be described in further detail with reference to the embodiments:

[0054] Referring to Figure 1, a fault early warning method for a refrigeration system based on an AI model includes the following steps:

[0055] S1 collects multiple status data of the refrigeration system, equipment design parameters, and maintenance work order records;

[0056] In this embodiment, the status data can be referenced in Table 1 below. By collecting some or all of these parameters, multiple operating parameters or status parameters of the refrigeration system can be obtained. Equipment design parameters can be directly obtained from the refrigeration system's nameplate, instruction manual, etc. The maintenance work order record contains historical fault records of the refrigeration system, covering the fault time period, faulty equipment, fault type, handling steps, and step costs.

[0057] Table 1, Example of Status Data

[0058]

[0059] S2, the state data is dynamically decoupled to remove the influence of environmental temperature and load factors, and the compensated state data is obtained.

[0060] In fault early warning systems for industrial equipment, environmental factors (such as ambient temperature) and load factors (such as equipment load rate) have a significant impact on the operating status of the equipment. These factors may cause fluctuations in certain parameters of the equipment, even if the equipment itself has not failed. Therefore, removing the influence of these factors can restore the status data to its original state, providing a baseline for subsequent steps.

[0061] S3, perform feature cross-fusion of the compensated status data, the equipment design parameters, and the maintenance work order records to generate comprehensive health status features;

[0062] In this embodiment, the compensated state data, the equipment design parameters, and the maintenance work order records are simultaneously considered through feature cross-fusion, and new features are generated through fusion. These features help to more comprehensively assess the health status of the equipment. Furthermore, by fusing the maintenance work order records, this embodiment can provide historical background information for the comprehensive health status features by incorporating parameters such as fault type, fault time, and fault urgency in the maintenance work orders, helping subsequent AI models understand the long-term operating status of the equipment and its possible fault modes. The compensated state data, the equipment design parameters, and the maintenance work order records are fused to form a comprehensive feature vector. These features may include: fault mechanism features, such as changes in the equipment's operating temperature and pressure, which can reflect the health status and potential faults of the equipment; and equipment performance features, such as the condensing efficiency index, system performance coefficient (SCOP), and chiller performance coefficient (COP), which can accurately reflect the actual operating conditions of the equipment and help predict equipment faults.

[0063] S4. Based on the comprehensive health status characteristics, establish an FMEA failure mode identification model, an FTA fault tree construction model, and an ETA event tree evolution model for the refrigeration system. Quantify the failure risk priority of the refrigeration system through the FMEA failure mode identification model, locate the underlying cause and probability of failure of the refrigeration system through the FTA fault tree construction model, and predict the failure evolution path of the refrigeration system through the ETA event tree evolution model.

[0064] In this embodiment, three models are generated based on the comprehensive health status characteristics: the FMEA (Failure Mode and Effects Analysis) model, the FTA (Fault Tree Acquisition) model, and the ETA (Event Tree Evolution) model. The FMEA model is a systematic failure mode identification and analysis method used to identify potential failure modes, assess their impact, and determine their severity, frequency, and detectability to formulate response strategies. By identifying potential failure modes of equipment, its Risk Priority Number (RPN) is quantified, providing a basis for subsequent fault warnings. The FTA model is a graphical logical analysis method used to analyze and identify the root causes of system failures. By constructing a fault tree, the occurrence of failures is decomposed layer by layer according to logical relationships to find the underlying causes leading to top-level failure events. Through fault tree analysis, failure events are decomposed layer by layer to find the root causes of system-level failures, helping to optimize system design and equipment maintenance. The ETA model is an event tree-based analysis method used to simulate the evolution process of failures, helping to predict the development path of failures. By simulating the failure evolution path, it predicts the possible development of equipment failures, provides early warnings, and helps maintenance personnel take timely measures.

[0065] By establishing these three models, we can obtain various failure modes, failure probabilities, failure evolution paths, top-level manifestations and underlying causes of failures in the refrigeration system, providing basic data for subsequent AI model training.

[0066] S5. The AI ​​model is trained by the comprehensive health status characteristics, the failure risk priority number and failure type output by the FMEA failure mode identification model, the underlying cause and probability of failure output by the FTA fault tree construction model, and the failure evolution path output by the ETA event tree evolution model, to obtain the refrigeration system health prediction AI model.

[0067] The core objective of this step is to generate a model capable of calculating the health of the equipment's refrigeration system by comprehensively analyzing various data from the equipment, thus providing a basis for maintenance decisions. Health status assessment includes analyzing the equipment's current operating status, potential faults, and historical data, and quantifying the equipment's health status by generating an equipment health index through an AI model. In this embodiment, the output data from three models—FMEA (Failure Mode and Effects Analysis), FTA (Fault Tree Acquisition), and ETA (Event Tree Evolution)—are all used to train the AI ​​model. The AI ​​model can then identify the potential correlations between the data output from these three models, linking the refrigeration system's health with various fault conditions, fault probabilities, fault evolution paths, and fault causes, thereby obtaining a corresponding AI model for predicting the health of the refrigeration system.

[0068] S6, based on the refrigeration system health prediction AI model, performs a health status assessment on the refrigeration system status data, equipment design parameters, and maintenance work order records, and generates an equipment health index;

[0069] This step uses the status data of the refrigeration system, equipment design parameters, and maintenance work order records as input parameters for the refrigeration system health prediction AI model, thereby obtaining the corresponding equipment health index output by the refrigeration system health prediction AI model.

[0070] S7. Sort the fault risk priority numbers output by the FMEA failure mode identification model, associate the range of the equipment health index with the order of the fault risk priority numbers, determine the corresponding fault risk priority numbers according to the equipment health index, and generate the priority of maintenance tasks and maintenance decisions.

[0071] In this step, the FMEA (Failure Mode and Effects Analysis) model can output multiple failure modes and a risk priority value for each failure mode. This value reflects the severity of each failure mode and is used to rank different failure modes and determine their processing priority. Furthermore, in S5, the trained refrigeration system health prediction AI model can find potential correlations between equipment health indices and multiple failure modes. Equipment with lower health indices indicates increased failure risk and may exhibit failure modes with higher risk priority values, requiring priority processing.

[0072] Furthermore, as mentioned above, the corresponding failure mode associated with the equipment can be obtained through the equipment health index. At this time, the underlying cause and evolution path of the failure mode can be obtained from the FTA fault tree construction model and the ETA event tree evolution model. Based on the underlying cause and evolution path of the failure, the maintenance decision of the failure can be determined, such as immediate handling or periodic handling.

[0073] Furthermore, in S1, after collecting multiple state data of the refrigeration system, the following is performed: predicting and filling in missing or abnormal values ​​in the multiple state data.

[0074] In this embodiment, during data acquisition, missing and outlier values ​​may exist in the raw data due to sensor malfunctions, communication interruptions, or environmental interference. To address the missing value problem, this application employs a random forest algorithm for imputation. This method predicts missing values ​​by constructing multiple decision tree models, thereby improving data integrity. For outlier detection and handling, a method based on the 3σ criterion is used. This assumes the data follows a normal distribution, considers data exceeding the mean ± 3 standard deviations as outliers, and corrects them by imputing with the median or deleting outlier samples.

[0075] For example, the chiller parameters are obtained as follows:

[0076] Raw condenser outlet temperature data: [34.2, 102.5 (interference), 34.4, NaN, 34.5];

[0077] After prediction imputation, we get: [34.2, 34.3 (median filtering), 34.4, 34.45 (interpolation), 34.5].

[0078] S2, the multiple state data are dynamically decoupled to remove the influence of environmental and load factors, resulting in compensated state data including:

[0079] S2.1, Establish a multiple linear regression model, and train the multiple linear regression model using ambient temperature, load rate, and the state data as inputs;

[0080] S2.2, The residual terms in the trained multiple linear regression model are used as compensation to adjust the state data to obtain compensated state data.

[0081] In this embodiment, a multiple linear regression model is used to describe the relationship between environmental factors, load factors, and equipment parameters. The multiple linear regression model can be expressed as follows:

[0082] ;

[0083] Where Tenv is the ambient temperature, Loadratio is the load ratio, and ΔToriginal is the original measurement value. It is an error term. and These are the regression coefficients. By training this multiple linear regression model with a large amount of raw data, we can obtain... At this point, the residual characteristics can be represented, through... Compensating the status data yields the corresponding compensated status data. The core purpose of calculating the compensated status data is to isolate the equipment's operating status from external factors, ensuring that abnormal equipment signals can be correctly identified. Based on this, the system adjusts the failure mode thresholds according to the equipment's health status, avoiding false alarms or missed alarms caused by environmental changes or load fluctuations.

[0084] Further, in step S3, the compensated status data, the equipment design parameters, and the maintenance work order records are cross-fused to generate comprehensive health status features, including:

[0085] S3.1, The compensated state data and the equipment design parameters are fused to generate a deviation fusion feature with physical meaning;

[0086] S3.2, the compensated status data and the maintenance work order record are fused to generate fault fusion features, and each fault fusion feature is labeled according to the fault type in the maintenance work order record;

[0087] S3.3, assign corresponding weights to the deviation fusion feature and the fault fusion feature, and then perform weighted fusion to generate the comprehensive health status feature.

[0088] In this step, the compensated state data of the equipment (such as the equipment's temperature, pressure, flow rate, etc.) are combined with the equipment's design parameters (such as rated power, safety boundary, system design parameters, etc.) to generate a fusion feature that can reflect the deviation between the equipment design and actual operation. Each fusion feature can correspond to a clear physical meaning, as shown in Table 2.

[0089] Table 2. Examples of fusion between compensated state data and equipment design parameters

[0090]

[0091] The compensated status data of the equipment (such as parameters such as temperature, pressure, and flow rate) is combined with the equipment's maintenance work order records (such as fault history, fault type, and maintenance cost information) to generate a fusion feature that reflects the health status and potential faults of the equipment. This fusion can identify a segment of the compensated status data, label it with the fault type, and find the corresponding timeline, fault verification degree, etc.

[0092] Next, the deviation fusion feature and the fault fusion feature are fused together to integrate multi-dimensional information from the current state of the equipment, historical fault records, and equipment design specifications. Furthermore, the changes in the physical meaning represented by multiple fusion features corresponding to the fault type can be found above. At this point, each change is assigned a corresponding weight based on its impact on the system health status. Simultaneously, expert experience (subjective weight) and statistical data patterns (objective weight) are comprehensively considered. For example, for a motor wear fault, the weight of the change in the condensing efficiency index is 0.1, and the weight of the change in the system performance coefficient is 0.5, etc. These weighted sums are then used to obtain the corresponding comprehensive health status feature.

[0093] Furthermore, the fusion of the compensated status data and the maintenance work order record includes:

[0094] The compensated status data and the maintenance work order record are time-aligned using a dynamic time warping algorithm.

[0095] The fault fusion feature is obtained by fusing the compensated state data within a preset time range before and after the fault time point in the maintenance work order record, and the fault type recorded in the maintenance work order record.

[0096] In this step, aligning the time series data with the maintenance logs is the first step in the fusion process, ensuring all relevant data have consistent timestamps. This is achieved through a dynamic time warping algorithm, which handles time deviations caused by inconsistent sampling frequencies of different devices or sensors, ensuring time alignment of data from different sources. Once the time series data and maintenance work order records are aligned, the equipment state characteristics before and after the fault can be extracted. Then, based on the time point in the maintenance work order record, compensated state data for a period before and after it is extracted. For example, parameters such as condenser outlet temperature, exhaust pressure, and suction pressure typically change within 10 minutes before and after condenser scaling. By analyzing the trends of these parameters, potential fault modes can be identified. By calculating the rate of change of features within a certain time window before and after the fault, it can be determined whether the equipment is malfunctioning. Simultaneously, the fault type and severity in the maintenance work order records can be labeled with the data within that time window. Effective fusion of feature changes before and after the fault not only helps identify the moment of the fault but also allows for the prediction of future potential faults based on feature changes before the fault occurs.

[0097] Furthermore, following S3.3, it includes:

[0098] S3.4, dimensionality reduction is performed on the comprehensive health status characteristics, and the principal components with the most influence on the health assessment of the refrigeration system are selected using the PCA method.

[0099] In this step, the principal components with a cumulative contribution rate of over 80% can be selected as the fused features by using the PCA method, thereby extracting the principal components, reducing feature redundancy, and improving model training efficiency.

[0100] Further, step S4 includes:

[0101] Based on the overall health status, potential failure modes of the refrigeration system are identified using the FMEA failure mode identification model. Each failure mode is assigned a severity, frequency of occurrence, and detectability. The failure risk priority number is calculated using the following formula:

[0102] ;

[0103] Based on the failure modes output by the FMEA failure mode identification model, an ETA event tree evolution model is constructed using logic gates to analyze top-level failure events, locate underlying causes of failures, and obtain the probability of occurrence of underlying causes of failures.

[0104] Based on the failure modes output by the FMEA failure mode identification model and the underlying causes and probabilities of failure output by the FTA fault tree construction model, an ETA event tree evolution model is constructed to simulate the possible paths of failure in the refrigeration system and predict the failure evolution path of the refrigeration system through the possible paths.

[0105] In this step, three models need to be trained.

[0106] Based on expert experience and the overall health status, AI training is used to identify equipment failure modes. Taking a central air conditioning system as an example, failure modes such as compressor overheating and condenser scaling may be identified. Then, the severity (S), frequency (O), and detectability (D) of these failure modes are quantified, and the failure risk priority number is obtained by using RPN=S×O×D, as shown in Table 3.

[0107] Table 3, Example Table of Fault Risk Priority Numbers.

[0108]

[0109] Based on the failure modes identified by FMEA, a fault tree model is constructed to locate the underlying causes. In the fault tree model, bottom events are connected to intermediate events via AND or OR gates, and intermediate events are connected to the top event via OR gates. The probability of occurrence is indicated between each branch. Subsequent AI training and model iteration can adjust the probability of occurrence to optimize the model and obtain the top, intermediate, and bottom events corresponding to the fault type. Taking the failure of refrigeration system efficiency reduction in a central air conditioning system as an example, the data structure can be seen in Figure 2. The method for event determination can be seen in Table 4.

[0110] Table 4. Fault tree model fault type events.

[0111]

[0112] The FMEA failure mode identification model can output various failure types and their evolution paths. For example, uncleaned filter → compressor overload probability increases by 65% ​​within 30 days. Based on the failure modes identified by FMEA, an event tree is constructed to simulate the failure evolution process, resulting in the following path for predicting failure evolution:

[0113] Initial event: The COP value of the refrigeration unit is 15% lower than the design value;

[0114] Possible development paths:

[0115] Decreased condenser efficiency → Insufficient cooling water flow → Loose fan belt or scaling in cooling tower;

[0116] Hydraulic system imbalance → abnormal valve opening → electric actuator malfunction or control signal error;

[0117] Abnormal sensor data → Sensor malfunction → Physical damage or environmental interference;

[0118] Termination condition: When a critical parameter is detected to exceed the safety threshold, a maintenance alarm is triggered.

[0119] Further, step S5 includes:

[0120] S5.1, By using the comprehensive health status characteristics described in the history, the time-series prediction model is trained to obtain the future operating parameter prediction model, which is used to predict the future operating parameters of the refrigeration system;

[0121] S5.2, The supervised learning model is trained using the comprehensive health status features and the fault type output by the FMEA failure mode recognition model to obtain a fault probability prediction model, which is used to predict the fault type and its fault probability.

[0122] S5.3, Based on the future operating parameters predicted by the future operating parameter prediction model and the failure type probability predicted by the failure probability prediction model, establish a refrigeration system health prediction AI model.

[0123] In this step, the supervised learning model aims to predict the probability or mode of equipment failure based on historical data. A random forest algorithm is used as the failure probability prediction model to build a classification model, predicting whether the equipment is likely to fail. Model inputs include features such as pressure difference, temperature difference, and flow balance, and the output is various failure types and their probabilities. The time-series prediction model analyzes the trend of equipment operating data over time and predicts future equipment status. An LSTM neural network is used to build a future operating parameter prediction model, predicting the changing trends of key parameters. The model input is historical time-series data, and the output is the predicted values ​​of future operating parameters.

[0124] Then, based on the parameters output by the failure probability prediction model and the future operating parameter prediction model, an AI model for evaluating the health prediction of the refrigeration system is established.

[0125] Furthermore, establishing an AI model for predicting the health of refrigeration systems includes:

[0126] Based on the deviation between the future operating parameters predicted by the future operating parameter prediction model and the equipment design parameters, the parameter deviation health of the refrigeration system is calculated.

[0127] The equipment aging health is calculated based on the failure type probability predicted by the failure probability prediction model and the comprehensive health status characteristics.

[0128] Based on the deviation between the energy efficiency index corresponding to the future operating parameters predicted by the future operating parameter prediction model and the energy efficiency index of the equipment design, the energy efficiency deviation health of the refrigeration system is calculated.

[0129] The health status of the parameter deviation, the health status of the equipment aging, and the health status of the energy efficiency deviation are added together according to their corresponding weights to obtain the AI ​​model for predicting the health of the refrigeration system.

[0130] In this embodiment, the parameter deviation health score HV1 of the refrigeration system can be used to measure the deviation between the equipment's operating parameters and design parameters. Equipment operating data (such as temperature, pressure, flow rate, etc.) may be affected by the external environment or equipment aging. Therefore, HV1 reflects the equipment's health status by calculating the deviation between the actual operating parameters and the design standard values. The energy efficiency deviation health score HV3 reflects the equipment's energy efficiency, particularly the deviation from the equipment's design values ​​(such as COP, SCOP, etc.). Poor energy efficiency usually indicates potential equipment malfunctions or performance problems. For example, by obtaining the ratio of the equipment's output cooling capacity to its energy consumption as the COP value, the deviation from the design target is calculated, thus obtaining the energy efficiency deviation health score HV3. The equipment aging health score HV2 mainly measures the degree of aging during equipment use, i.e., wear or performance degradation of the equipment during long-term operation. Aging health considers factors such as the equipment's service life, historical fault records, and equipment maintenance frequency. Maintenance records, the number of faults, and the type of faults can also be used to assess aging. Frequent faults or equipment requiring extensive repairs may be a signal of equipment aging. HV3 and HV3 can be classified or assigned based on expert experience. Then, the health status of the refrigeration system is calculated using the following formula. :

[0131] The weights are 0.3, 0.2, and 0.5, respectively.

[0132] Further, in step S7, associating the range of the equipment health index with the order of the fault risk priority numbers includes:

[0133] The equipment health index and the fault risk priority number are normalized, and the normalized fault risk priority number is mapped to the normalized equipment health index in descending order.

[0134] The above steps yielded the health status of the refrigeration system. After normalizing the aforementioned failure risk priority number, the health of the refrigeration system can be assessed. The fault risk priority number is mapped to a value between 0 and 1, due to the health of the refrigeration system. This can be expressed as a higher numerical value indicating better health of the refrigeration system, while the fault risk priority number indicates a higher risk level. Therefore, the health of the refrigeration system can be assessed based on its overall health. The descending order is mapped one-to-one with the fault risk priority number, thus obtaining each fault type and its fault risk priority number.

[0135] For example, a leaking pressure relief valve is one of the highest priority fault risks and requires immediate attention. It has the highest priority. On the other hand, problems such as scale buildup generally do not require immediate attention and have a lower priority.

[0136] Further, in step S7, the priority of generating maintenance tasks includes: prioritizing the fault risk priority number and obtaining the corresponding priority based on the fault risk priority number corresponding to the equipment health index;

[0137] Maintenance decisions for generating maintenance tasks include: preventive maintenance for fault types with low probability of failure but high impact, corrective maintenance for faults with high probability of failure and high impact, and optimization maintenance for maintenance that does not affect safety but can improve energy efficiency.

[0138] This step further considers maintenance decisions for equipment maintenance tasks.

[0139] Preventive maintenance targets faults with a low probability of failure but a significant impact. For example, equipment may have a low health index, but the probability of failure is low. Preventive maintenance typically does not require immediate repair, but it is recommended to regularly inspect or replace parts to avoid future equipment failures.

[0140] Corrective maintenance is for faults with a high probability of failure and significant impact, such as when the equipment's health index is very low and the RPN value is high. In this case, the equipment's health condition has deteriorated, and the system will recommend immediate repair. For example, if the equipment's compressor oil pressure is abnormal and the health index is very low, the system will suggest immediate corrective maintenance to avoid a complete equipment shutdown.

[0141] Optimization maintenance addresses equipment energy efficiency issues, such as low energy efficiency or decreased system performance, but does not immediately lead to equipment failure. For example, if equipment has a high health index but exhibits certain energy efficiency problems (such as low condenser heat exchange efficiency), optimization maintenance might be recommended to improve system efficiency and save energy.

[0142] This application generates maintenance task priorities and maintenance decisions by associating the range of equipment health indices with the order of failure risk priority numbers. Combining the ranking of health indices and failure risk priority numbers ensures that both the equipment's health status and the risk of failure modes are considered simultaneously.

[0143] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0144] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in one or more blocks of the flowchart illustrations and / or one or more blocks of the block diagrams.

[0145] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means that implement the functions specified in one or more flowcharts and / or one or more block diagrams.

[0146] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0147] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms should not be construed as necessarily referring to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

Claims

1. A method for early warning of refrigeration system faults based on an AI model, characterized in that: Includes the following steps: S1. Collect multiple status data of the refrigeration system, equipment design parameters, and maintenance work order records; S2. Dynamically decouple the multiple status data to remove the influence of environmental temperature and load factors, obtaining compensated status data; S3. Perform feature cross-fusion of the compensated status data, equipment design parameters, and maintenance work order records to generate comprehensive health status features; S4. Based on the comprehensive health status features, establish an FMEA failure mode identification model, an FTA fault tree construction model, and an ETA event tree evolution model for the refrigeration system. The FMEA failure mode identification model quantifies the failure risk priority of the refrigeration system, the FTA fault tree construction model locates the underlying causes and probabilities of failures in the refrigeration system, and the ETA event tree evolution model predicts the failure status of the refrigeration system. Fault evolution path; S5, using the comprehensive health status features, the fault risk priority number and fault type output by the FMEA failure mode recognition model, the underlying causes and occurrence probabilities of the faults output by the FTA fault tree construction model, and the fault evolution path output by the ETA event tree evolution model, train the AI ​​model to obtain the refrigeration system health prediction AI model; including: S5.1, using the historical comprehensive health status features, train the time series prediction model to obtain the future operating parameter prediction model, used to predict the future operating parameters of the refrigeration system; S5.2, using the comprehensive health status features and the fault types output by the FMEA failure mode recognition model, train the supervised learning model to obtain the fault probability prediction model, used to predict the fault type and its fault probability; S5.

3. Based on the future operating parameters predicted by the future operating parameter prediction model and the failure type probabilities predicted by the failure probability prediction model, establish a refrigeration system health prediction AI model. Establishing the refrigeration system health prediction AI model includes: calculating the parameter deviation health of the refrigeration system based on the deviation between the future operating parameters predicted by the future operating parameter prediction model and the equipment design parameters; calculating the equipment aging health based on the failure type probabilities predicted by the failure probability prediction model and the comprehensive health status characteristics; and calculating the refrigeration system health based on the deviation between the energy efficiency indicators corresponding to the future operating parameters predicted by the future operating parameter prediction model and the energy efficiency indicators designed for the equipment. The system's energy efficiency deviation health status is assessed. The parameter deviation health status, equipment aging health status, and energy efficiency deviation health status are weighted together to obtain the refrigeration system health prediction AI model. S6: Based on the refrigeration system health prediction AI model, the system's status data, equipment design parameters, and maintenance work order records are evaluated to generate an equipment health index. S7: The fault risk priority numbers output by the FMEA failure mode recognition model are sorted. The range of the equipment health index is correlated with the order of the fault risk priority numbers. Based on the equipment health index, the corresponding fault risk priority numbers are determined, generating maintenance task priorities and maintenance decisions.

2. The method for early warning of refrigeration system faults based on an AI model according to claim 1, characterized in that: In S1, after collecting multiple state data of the refrigeration system, the following steps are performed: Predicting and filling in missing or abnormal values ​​in the multiple state data; In S2, the multiple state data are dynamically decoupled to remove the influence of environmental and load factors, resulting in compensated state data, including: S2.1, establishing a multiple linear regression model, and training the model with ambient temperature, load rate, and the state data as inputs; S2.2, using the residual terms in the trained multiple linear regression model as compensation to adjust the multiple state data, resulting in compensated state data.

3. The method for early warning of refrigeration system faults based on an AI model according to claim 1, characterized in that: S3, performing feature cross-fusion of the compensated state data, the equipment design parameters, and the maintenance work order records to generate a comprehensive health status feature includes: S3.1, fusing the compensated state data and the equipment design parameters to generate a deviation fusion feature with physical meaning; S3.2, fusing the compensated state data and the maintenance work order records to generate a fault fusion feature, and labeling each fault fusion feature according to the fault type in the maintenance work order record; S3.3, assigning corresponding weights to the deviation fusion feature and the fault fusion feature and then weighting and fusing them to generate the comprehensive health status feature.

4. The method for early warning of refrigeration system faults based on an AI model according to claim 3, characterized in that: The fusion of the compensated state data and the maintenance work order record includes: using a dynamic time warping algorithm to align the compensated state data and the maintenance work order record in time; fusing the compensated state data within a preset time range before and after the fault time point in the maintenance work order record, as well as the fault type recorded in the maintenance work order record, to obtain the fault fusion feature.

5. The method for early warning of refrigeration system faults based on an AI model according to claim 4, characterized in that: Following S3.3, the process includes: S3.4, which involves dimensionality reduction of the comprehensive health status characteristics and selection of the principal components that have the greatest impact on the health assessment of the refrigeration system using the PCA method.

6. The method for early warning of refrigeration system faults based on an AI model according to claim 1, characterized in that: Step S4 includes: based on the comprehensive health status, identifying potential failure modes of the refrigeration system using the FMEA failure mode identification model, and assigning severity, frequency of occurrence, and detectability to each failure mode, and calculating the failure risk priority number using the following formula: Based on the failure modes output by the FMEA failure mode identification model, an ETA event tree evolution model is constructed using logic gates to analyze top-level failure events, locate underlying causes of failures, and obtain the probability of occurrence of underlying causes of failures. Based on the failure modes output by the FMEA failure mode identification model, the underlying causes of failures output by the FTA fault tree construction model, and their probability of occurrence, an ETA event tree evolution model is constructed to simulate possible failure paths of the refrigeration system and predict the failure evolution path of the refrigeration system through these possible paths.

7. The method for early warning of refrigeration system faults based on an AI model according to claim 1, characterized in that: In step S7, associating the range of the equipment health index with the order of the fault risk priority number includes: normalizing the equipment health index and the fault risk priority number, and mapping the normalized fault risk priority number to the normalized equipment health index in descending order.

8. The method for early warning of refrigeration system faults based on an AI model according to claim 7, characterized in that: In step S7, the priority of generating maintenance tasks includes: prioritizing the fault risk priority number and obtaining the corresponding priority based on the fault risk priority number corresponding to the equipment health index; the maintenance decision for generating maintenance tasks includes: performing preventive maintenance for fault types with low fault probability but high impact, performing corrective maintenance for faults with high fault probability and high impact, and performing optimization maintenance for maintenance that does not affect safety but can improve energy efficiency.

Images