Air conditioning chiller fault diagnosis method and system based on multimodal data fusion

Abstract

The present invention relates to the technical field of air-conditioning chillers, and provides an air-conditioning chiller fault diagnosis method and system based on multimodal data fusion. The method includes: fuzzy processing each variable in internal data, sound data and vibration data to obtain internal features, sound features and vibration features; cross-fusing the internal features, sound features and vibration features in pairs to obtain a first fusion feature, a second fusion feature and a third fusion feature; respectively calculating the first similarity of the first fusion feature and the second fusion feature, and calculating the second similarity of the second fusion feature and the third fusion feature, using the first similarity as the weight of the first fusion feature, and using the second similarity as the weight of the third fusion feature; fusing the first fusion feature with weights, the third fusion feature and the second fusion feature to obtain a multimodal fusion feature; using a fault diagnosis model to obtain a fault diagnosis result of the air-conditioning chiller.

Description

Technical Field

[0001] The present invention relates to the technical field of air-conditioning chillers, and in particular to an air-conditioning chiller fault diagnosis method and system based on multi-modal data fusion. Background Art

[0002] The statements in this section merely provide background information related to the present invention and do not necessarily constitute prior art.

[0003] Traditional fault diagnosis methods for air-conditioning chillers mainly rely on regular inspections and simple threshold monitoring, but due to the wide variety of equipment and complex failure modes, these methods often fail to provide timely and accurate fault diagnosis. For example, regular inspections are difficult to cover all possible fault conditions, and long inspection intervals may lead to the accumulation of hidden dangers; simple threshold detection methods may not be able to capture subtle changes in multimodal data, resulting in insufficient sensitivity in diagnosis. This limitation may lead to delayed discovery or misdiagnosis of equipment failures, thereby increasing maintenance costs and equipment downtime. Summary of the invention

[0004] In order to solve the technical problems existing in the above-mentioned background technology, the present invention provides an air-conditioning chiller fault diagnosis method and system based on multimodal data fusion. The present invention provides efficient fault detection capabilities, can instantly identify and respond to equipment problems, and reduce downtime and maintenance costs.

[0005] In order to achieve the above object, the present invention adopts the following technical solution:

[0006] A first aspect of the present invention provides an air-conditioning chiller fault diagnosis method based on multimodal data fusion.

[0007] The air conditioning chiller fault diagnosis method based on multimodal data fusion includes:

[0008] Acquire internal data inside the air-conditioning chiller, sound data outside the air-conditioning chiller, and vibration data outside the compressor; the internal data includes temperature data, pressure data, flow data, and current data;

[0009] Perform fuzzy processing on each variable in the internal data, sound data and vibration data to obtain internal features, sound features and vibration features;

[0010] The internal features, sound features and vibration features are cross-fused in pairs to obtain the first fusion feature, the second fusion feature and the third fusion feature; the first similarity between the first fusion feature and the second fusion feature is calculated respectively, and the second similarity between the second fusion feature and the third fusion feature is calculated, and the first similarity is used as the weight of the first fusion feature, and the second similarity is used as the weight of the third fusion feature; the first fusion feature and the third fusion feature with weights are fused with the second fusion feature to obtain a multimodal fusion feature;

[0011] Based on multimodal fusion features and the fault diagnosis model, the fault diagnosis results of the air-conditioning chiller are obtained.

[0012] Further, the internal features, the sound features, and the vibration features are cross-fused in pairs to obtain a first fusion feature, a second fusion feature, and a third fusion feature. The method includes:

[0013] Based on the internal features and vibration features, a multi-head attention mechanism is used to linearly map the internal features and vibration features to generate a first multi-dimensional attention weight; the first multi-dimensional attention weight is used to obtain the interaction features between the modalities, and the first fusion features of the internal data and the vibration data are output;

[0014] Based on the sound features and vibration features, a multi-head attention mechanism is used to linearly map the sound features and vibration features to generate a second multi-dimensional attention weight; the second multi-dimensional attention weight is used to obtain the interaction features between the modalities, and the second fusion features of the sound data and the vibration data are output;

[0015] Based on internal features and sound features, a multi-head attention mechanism is used to perform linear mapping on internal features and sound features to generate third-dimensional attention weights. The third-dimensional attention weights are used to obtain the interaction features between modalities and output the third fusion features of internal data and sound data.

[0016] Furthermore, each variable in the internal data is fuzzy processed to obtain internal characteristics. The method includes: defining fuzzy sets and membership functions for temperature data, pressure data, flow data and current data respectively, and associating temperature data, pressure data, flow data and current data to different fuzzy sets through membership functions; wherein one membership function corresponds to one fuzzy data set, and low-temperature membership function, medium-temperature membership function, high-temperature membership function, low-pressure membership function, medium-pressure membership function, high-pressure membership function, low-flow membership function, medium-flow membership function, high-flow membership function, low-current membership function, medium-current membership function and high-current membership function correspond to different fuzzy sets respectively.

[0017] Furthermore, the low temperature membership function is expressed by the following formula:

[0018]

[0019] in, The fuzzy set L corresponding to the temperature T The membership function, T min is the minimum temperature value in the low temperature range, L T It is the highest temperature value in the low temperature range or the lowest temperature value in the medium temperature range.

[0020] Furthermore, the medium temperature membership function is expressed by the following formula:

[0021]

[0022] in, The second fuzzy set M corresponding to temperature T The membership function, L T is the highest temperature in the low temperature range or the lowest temperature in the medium temperature range, M T is the middle value of the medium temperature range, H T It is the lowest temperature value in the high temperature range or the highest temperature value in the medium temperature range.

[0023] Furthermore, the high temperature membership function is expressed by the following formula:

[0024]

[0025] in, The fuzzy set H corresponding to temperature T The membership function, M T is the middle value of the medium temperature range, H T is the lowest temperature in the high temperature range or the highest temperature in the medium temperature range, T max It is the highest temperature value in the high temperature range.

[0026] Furthermore, each variable in the internal data is fuzzified to obtain internal features, and the method includes:

[0027] The actual value of each variable in the fuzzy set is converted into fuzzy membership, and the fuzzy membership of each variable is concatenated as a feature to obtain the internal feature.

[0028] Further, each variable in the sound data and the vibration data is fuzzy processed to obtain the sound feature and the vibration feature, and the method includes:

[0029] Define fuzzy sets and membership functions for sound data and vibration data respectively, and associate the sound data and vibration data to different fuzzy sets through the membership functions;

[0030] The actual value of each variable in the fuzzy set corresponding to the sound data is converted into a fuzzy membership degree, and the fuzzy membership degree of each variable is used as a feature for splicing to obtain the sound feature;

[0031] The actual value of the fuzzy set corresponding to the vibration data is converted into the fuzzy membership, which is the vibration feature.

[0032] Further, the temperature data includes: the refrigerant temperature at the condenser inlet, the refrigerant temperature at the condenser outlet, the refrigerant temperature at the evaporator inlet, the refrigerant temperature at the evaporator outlet, the refrigerant temperature of the compressor suction pipe, the refrigerant temperature of the compressor exhaust pipe, the refrigerant temperature at the front end of the throttling device, and the refrigerant temperature at the rear end of the throttling device;

[0033] The pressure data include: the refrigerant pressure at the condenser inlet, the refrigerant pressure at the condenser outlet, the refrigerant pressure at the evaporator inlet, the refrigerant pressure at the evaporator outlet, the refrigerant pressure at the compressor suction pipe, the refrigerant pressure at the compressor exhaust pipe, the refrigerant pressure at the front end of the throttling device, and the refrigerant pressure at the rear end of the throttling device;

[0034] The flow data includes: the refrigerant flow at the condenser inlet and the refrigerant flow at the evaporator inlet;

[0035] The current data includes: the working current of the compressor;

[0036] The sound data includes abnormal noise data outside the condenser, abnormal noise data outside the evaporator, abnormal noise data outside the compressor, and abnormal noise data outside the throttling device.

[0037] A second aspect of the present invention provides an air conditioning chiller fault diagnosis system based on multimodal data fusion.

[0038] An air-conditioning chiller fault diagnosis system based on multimodal data fusion comprises: a temperature sensor installed inside the air-conditioning chiller, a pressure sensor installed outside the air-conditioning chiller, a flow sensor installed outside the air-conditioning chiller, a current sensor installed in a compressor power supply line, a condenser microphone installed outside the air-conditioning chiller, a condenser accelerometer installed on a compressor housing, and a processor; the processor is connected to the temperature sensor, the pressure sensor, the flow sensor, the current sensor, the condenser microphone, and the condenser accelerometer;

[0039] A temperature sensor is used to collect temperature data, a pressure sensor is used to collect pressure data, a flow sensor is used to collect flow data, a current sensor is used to collect current data, a capacitor microphone is used to collect sound data, and a capacitive accelerometer is used to collect vibration data; internal data includes: temperature data, pressure data, flow data and current data;

[0040] The processor is used to perform fuzzy processing on each variable in the internal data, sound data and vibration data to obtain internal features, sound features and vibration features; cross-fuse the internal features, sound features and vibration features in pairs to obtain the first fusion feature, the second fusion feature and the third fusion feature; respectively calculate the first similarity of the first fusion feature and the second fusion feature, and calculate the second similarity of the second fusion feature and the third fusion feature, use the first similarity as the weight of the first fusion feature, and use the second similarity as the weight of the third fusion feature; fuse the first fusion feature, the third fusion feature and the second fusion feature with weights to obtain multimodal fusion features; based on the multimodal fusion features, adopt a fault diagnosis model to obtain the fault diagnosis result of the air-conditioning chiller.

[0041] Compared with the prior art, the present invention has the following beneficial effects:

[0042] (1) The present invention provides efficient fault detection capabilities, which can instantly identify and respond to equipment problems, reducing downtime and maintenance costs; it can automatically trigger alarms and record data, reducing manual intervention, and improving system operation efficiency and response speed.

[0043] (2) By integrating environmental data, vibration data and sound data, the system of the present invention can comprehensively analyze the equipment status and improve the accuracy and robustness of detection.

[0044] (3) The present invention utilizes a fuzzy module for feature extraction, which can handle uncertainty and fuzziness problems and improve the accuracy of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings in the specification, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0046] Figure 1 is a flow chart of an air-conditioning chiller fault diagnosis method based on multimodal data fusion shown in the present invention;

[0047] Figure 2 is a structural diagram of a multimodal data driven fault diagnosis module shown in the present invention;

[0048] Figure 3 is a network structure diagram of the multimodal data fusion process shown in the present invention;

[0049] Figure 4 It is a framework diagram of the air-conditioning chiller fault diagnosis method based on multimodal data fusion shown in the present invention. DETAILED DESCRIPTION

[0050] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.

[0051] It should be noted that the following detailed descriptions are all illustrative and intended to provide further explanation of the present invention. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present invention belongs.

[0052] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0053] It should be noted that the flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the methods and systems according to various embodiments of the present disclosure. It should be noted that each box in the flowchart or block diagram can represent a module, a program segment, or a part of a code, and the module, program segment, or a part of a code may include one or more executable instructions for implementing the logical functions specified in each embodiment. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, or they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the flowchart and / or block diagram, and the combination of boxes in the flowchart and / or block diagram can be implemented using a dedicated hardware-based system that performs a specified function or operation, or can be implemented using a combination of dedicated hardware and computer instructions.

[0054] Embodiment 1

[0055] like Figure 1 As shown, this embodiment provides an air conditioning chiller fault diagnosis method based on multimodal data fusion, including:

[0056] Acquire internal data inside the air-conditioning chiller, sound data outside the air-conditioning chiller, and vibration data outside the compressor; the internal data includes temperature data, pressure data, flow data, and current data;

[0057] Perform fuzzy processing on each variable in the internal data, sound data and vibration data to obtain internal features, sound features and vibration features;

[0058] The internal features, sound features and vibration features are cross-fused in pairs to obtain the first fusion feature, the second fusion feature and the third fusion feature; the first similarity between the first fusion feature and the second fusion feature is calculated respectively, and the second similarity between the second fusion feature and the third fusion feature is calculated, and the first similarity is used as the weight of the first fusion feature, and the second similarity is used as the weight of the third fusion feature; the first fusion feature and the third fusion feature with weights are fused with the second fusion feature to obtain a multimodal fusion feature;

[0059] Based on multimodal fusion features and the fault diagnosis model, the fault diagnosis results of the air-conditioning chiller are obtained.

[0060] The present invention provides efficient fault detection capabilities, can instantly identify and respond to equipment problems, reduce downtime and maintenance costs; can automatically trigger alarms and record data, reduce manual intervention, and improve system operation efficiency and response speed.

[0061] The internal data set is represented as E, and each day’s data contains 1440 time samples and 19 feature variables. Specifically, the data E on the i-th day i It can be represented as a 1440×19 matrix:

[0062]

[0063] in, Represents the value of the pth characteristic variable corresponding to the tth time point, i = 1, 2, ..., 30, t = 1, 2, ..., 1440, p = 1, 2, ..., 19.

[0064] The sound data set S consists of five feature vectors (i.e., the sound pressure levels in dB of five condenser microphones). The sound data S of the i-th day is i It can be represented as a 1440×5 matrix:

[0065] in, Represents the value of the pth sound feature variable corresponding to the tth time point, i=1,2,…,30, t=1,2,…,1440, p=1,2,3,4,5.

[0066] The vibration data set V consists of three characteristic variables (i.e., the acceleration of the X-axis, Y-axis, and Z-axis), and the acceleration values ​​of these three directions are recorded at each time point. The vibration data of the i-th day can be represented as a 1440×3 matrix:

[0067]

[0068] in, It represents the acceleration value on the X-axis at the t-th time point on the i-th day, i=1,2,…,30, t=1,2,…,1440.

[0069] The data collected at time t on day i is:

[0070]

[0071] right The corresponding air-conditioning chiller status is normal, condenser water flow reduction, evaporator water flow reduction, refrigerant leakage / insufficient, refrigerant excess, oil excess, condenser scaling, and refrigerant containing non-condensable gas, which are represented by 0-7 respectively.

[0072] in, It is one of eight states from 0 to 7.

[0073] like Figure 2 As shown, the internal data is input into the fuzzy module 1. The four input variables are: temperature (T), pressure (P), flow (F), and current (I). The corresponding fuzzy set and membership function are defined for each variable. Through the membership function, the temperature (T) is associated with the fuzzy set L T 、M T , H T ; Pressure (P) is associated with the fuzzy set L P 、M P , H P ; Flow (F) is associated with the fuzzy set L F 、M F , H F ; Current (I) is associated with the fuzzy set L I 、M I , H I . Where L T , L P , L F , L I Respectively represent low temperature, low pressure, low flow, and low current. T 、M P 、M F 、M I Respectively represent medium temperature, medium pressure, medium flow, and medium current. T , H P , H F , H I They respectively represent high temperature, high pressure, high flow, and high current.

[0074] The membership function selects trigonometric function, where the fuzzy set L corresponding to temperature (T) T 、M T and H TThe membership functions are defined as:

[0075]

[0076] in, The fuzzy set L corresponding to the temperature T The membership function, T min is the minimum temperature value in the low temperature range, L T It is the highest temperature value in the low temperature range or the lowest temperature value in the medium temperature range.

[0077]

[0078] in, The fuzzy set M corresponding to temperature T The membership function, L T is the highest temperature in the low temperature range or the lowest temperature in the medium temperature range, M T is the middle value of the medium temperature range, H T It is the lowest temperature value in the high temperature range or the highest temperature value in the medium temperature range.

[0079]

[0080] in, The fuzzy set H corresponding to temperature T The membership function of .

[0081] The fuzzy set membership functions of pressure (P), flow (F), and current (I) adopt the membership function similar to the above-mentioned temperature.

[0082] Temperature (T), pressure (P), flow (F), and current (I) are extracted through 19 sensors to obtain 19 feature vectors (including 8 temperature data, 8 pressure data, 2 flow data, and 1 current data), and each feature vector is associated with three fuzzy membership functions, ultimately obtaining 57 fuzzy extracted feature vectors.

[0083] Through the fuzzification process, the actual value of each input variable is converted into a fuzzy membership. The fuzzy membership of each variable is concatenated as a feature to obtain the internal feature vector:

[0084] F E =[T1,T2,…T8,P1,P2,…P8,F1,F2,I]

[0085] in,

[0086] The sound data and vibration data are input into fuzzy module 2 and fuzzy module 3 respectively to obtain the extracted feature vector, vibration feature F V and sound characteristics FS , and then perform multimodal data fusion.

[0087] (1) Cross-fusion between modes. Since the cross-fusion calculations under internal, vibration, and sound modes are the same, we take the internal and vibration feature fusion as an example and input the internal feature F E and vibration characteristics F V , based on the multi-head attention mechanism, linearly map the internal features and vibration features to generate multi-dimensional attention weights. Use the attention weights to capture the connection between the internal and vibration modal information and obtain the interaction features between the modalities. Output the first fusion feature F of the internal and vibration data EV .

[0088] (2) The first fusion feature F of the internal and vibration data is obtained EV , the second fusion feature F of vibration and sound data VS , the third fusion feature F of internal and sound data ES For the first fusion feature F EV and the third fusion feature F ES Both of them fuse the internal data. In order to balance the weight of the modality, the first fusion feature F is calculated separately. EV With the second fusion feature F VS , the third fusion feature F ES With the second fusion feature F VS The cosine similarity α and β of EV and the third fusion feature F ES The weight of the modality.

[0089] (3) The first fusion feature F with weight EV , the third fusion feature F ES With the second fusion feature F VS After further fusion, the final multimodal fusion feature F is obtained as:

[0090] F = LayerNorm(αF EV +βF ES +F VS )

[0091] Where α = cos(sim(F EV ,F VS )), β=cos(sim(F ES ,F VS )). The multimodal fusion feature F is input into the fault diagnosis model for model training.

[0092] The multimodal fusion feature F is input into the Fast-RCNN network model for training. During the training process, the model is optimized and the parameters are adjusted multiple times to build a fault diagnosis model with accurate fault diagnosis results. The specific steps of this module are as follows:

[0093] (1) The multimodal fusion feature F is divided into the training set F train and the test set F test , the training set F train Input into the fault diagnosis model (Fast-RCNN network model) for training.

[0094] (2) During the model training process, the cross entropy loss function and Adam optimization algorithm are used to optimize the model and adjust the parameters. The cross entropy loss function is calculated as:

[0095]

[0096] Among them, y is the true label, is the model prediction value, and n is the number of samples.

[0097] (3) After model training, input the test set F test Multimodal fusion data, the k-fold cross validation method is used to verify and evaluate the model. According to the evaluation results, the model structure and hyperparameters are further adjusted to obtain a fault diagnosis model with high accuracy.

[0098] First, unlabeled real-time data is collected through the data acquisition and annotation module, and real-time diagnosis is performed on the unlabeled real-time data. Fuzzy feature extraction and multimodal fusion are performed on the internal data, vibration data and sound data collected in real time to obtain multimodal fusion data. Finally, the multimodal fusion data is input into the fault diagnosis model for real-time fault diagnosis to obtain the diagnosis result.

[0099] The diagnostic results are saved as text data and uploaded to the database. If the system finds a fault, an alarm will be issued, and the operation and maintenance personnel will repair it and upload the repair records to the database; if no fault is detected, the test results will be uploaded to the database as regular maintenance data for later reference.

[0100] Embodiment 2

[0101] The present embodiment provides an air-conditioning chiller fault diagnosis system based on multimodal data fusion, comprising: a temperature sensor installed inside the air-conditioning chiller, a pressure sensor installed outside the air-conditioning chiller, a flow sensor installed outside the air-conditioning chiller, a current sensor installed in a compressor power supply line, a condenser microphone installed outside the air-conditioning chiller, a condenser accelerometer installed on a compressor housing, and a processor; the processor is connected to the temperature sensor, the pressure sensor, the flow sensor, the current sensor, the condenser microphone, and the condenser accelerometer;

[0102] A temperature sensor is used to collect temperature data, a pressure sensor is used to collect pressure data, a flow sensor is used to collect flow data, a current sensor is used to collect current data, a capacitor microphone is used to collect sound data, and a capacitive accelerometer is used to collect vibration data; internal data includes: temperature data, pressure data, flow data and current data;

[0103] The processor is used to perform fuzzy processing on each variable in the internal data, sound data and vibration data to obtain internal features, sound features and vibration features; cross-fuse the internal features, sound features and vibration features in pairs to obtain the first fusion feature, the second fusion feature and the third fusion feature; respectively calculate the first similarity of the first fusion feature and the second fusion feature, and calculate the second similarity of the second fusion feature and the third fusion feature, use the first similarity as the weight of the first fusion feature, and use the second similarity as the weight of the third fusion feature; fuse the first fusion feature, the third fusion feature and the second fusion feature with weights to obtain multimodal fusion features; based on the multimodal fusion features, adopt a fault diagnosis model to obtain the fault diagnosis result of the air-conditioning chiller.

[0104] The processor includes a multi-module data fusion module, a fault diagnosis model building module and a real-time diagnosis module. The system collects and processes multi-modal data of the air-conditioning chiller in real time, and uses fuzzy feature extraction and multi-modal data fusion to perform intelligent fault diagnosis and fault type judgment. In addition, the system also includes: a multi-modal data collection and annotation module. The following is a detailed description of this embodiment. Figure 4 As shown:

[0105] Module 1: Multimodal data collection and annotation module

[0106] This module is responsible for collecting data from different sensors installed inside and outside the air-conditioning chiller. Temperature (T), pressure (P), flow (F), and current (I) sensors are installed inside the four core parts of the air-conditioning chiller: evaporator, condenser, compressor, and throttling device to measure internal data; a condenser microphone is installed outside the air-conditioning chiller to measure the sound wave (S) signal; a capacitive accelerometer is installed outside the compressor to measure the vibration (V) signal. The specific sensor installation method is as follows:

[0107] Temperature sensors are installed at the inlet and outlet of the condenser to measure the refrigerant temperature, and pressure sensors are installed to measure the refrigerant pressure. A flow sensor is installed at the inlet of the condenser to measure the refrigerant flow. A condenser microphone is installed on the pipeline near the condenser and on the outside of the fan to monitor abnormal noise.

[0108] Temperature sensors are installed at the inlet and outlet of the evaporator to measure the refrigerant temperature, and pressure sensors are installed to measure the refrigerant pressure. A flow sensor is installed at the inlet of the evaporator to measure the refrigerant flow. A condenser microphone is installed on the outside of the pipeline near the evaporator to monitor abnormal noise.

[0109] Temperature sensors are installed on the suction pipe and exhaust pipe of the compressor to measure the refrigerant temperature, and pressure sensors are installed to measure the refrigerant pressure. Current sensors are installed in the power supply circuit of the compressor to measure the working current. Capacitive accelerometers and condenser microphones are installed on the compressor casing to monitor vibration and abnormal noise.

[0110] Temperature sensors are installed before and after the throttling device to measure the refrigerant temperature, and pressure sensors are installed to measure the refrigerant pressure. A condenser microphone is installed outside the expansion valve of the throttling device to monitor abnormal noise.

[0111] Data collection was done on a daily basis with a sampling interval of one minute, and a total of 30 days of data were collected.

[0112] The internal data set is denoted as E, and each day’s data contains 1440 time samples and 19 feature variables. Specifically, the data E on the i-th day i It can be represented as a 1440×19 matrix:

[0113]

[0114] in, Represents the value of the pth characteristic variable corresponding to the tth time point, i = 1, 2, ..., 30, t = 1, 2, ..., 1440, p = 1, 2, ..., 19.

[0115] The sound data set S consists of five feature vectors (i.e., the sound pressure levels in dB of five condenser microphones). The sound data S of the i-th day is iIt can be represented as a 1440×5 matrix:

[0116] in, Represents the value of the pth sound feature variable corresponding to the tth time point, i=1,2,…,30, t=1,2,…,1440, p=1,2,3,4,5.

[0117] The vibration data set V consists of three characteristic variables (i.e., the acceleration of the X-axis, Y-axis, and Z-axis), and the acceleration values ​​of these three directions are recorded at each time point. The vibration data of the i-th day can be represented as a 1440×3 matrix:

[0118]

[0119] in, It represents the acceleration value on the X-axis at the t-th time point on the i-th day, i=1,2,…,30, t=1,2,…,1440.

[0120] The data collected at time t on day i is:

[0121]

[0122] right The corresponding air-conditioning chiller status is normal, condenser water flow reduction, evaporator water flow reduction, refrigerant leakage / insufficient, refrigerant excess, oil excess, condenser scaling, and refrigerant containing non-condensable gas, which are represented by 0-7 respectively.

[0123] in, It is one of eight states from 0 to 7.

[0124] Module 2: Multimodal data feature extraction and fusion module

[0125] (1) Multimodal data fuzzy feature extraction method, such as Figure 2 shown.

[0126] Internal data is input into fuzzy module 1. The four input variables are: temperature (T), pressure (P), flow (F), and current (I). The corresponding fuzzy set and membership function are defined for each variable. Through the membership function, the temperature (T) is associated with the fuzzy set L T 、M T , H T ; Pressure (P) is associated with the fuzzy set L P 、M P , H P ; Flow (F) is associated with the fuzzy set L F 、M F , HF ; Current (I) is associated with the fuzzy set L I 、M I , H I . Where L T , L P , L F , L I Respectively represent low temperature, low pressure, low flow, and low current. T 、M P 、M F 、M I Respectively represent medium temperature, medium pressure, medium flow, and medium current. T , H P , H F , H I They respectively represent high temperature, high pressure, high flow, and high current.

[0127] The membership function selects trigonometric function, where the fuzzy set L corresponding to temperature (T) T 、M T and H T The membership functions are defined as:

[0128]

[0129] in, The fuzzy set L corresponding to the temperature T The membership function, T min is the minimum temperature value in the low temperature range, L T It is the highest temperature value in the low temperature range or the lowest temperature value in the medium temperature range.

[0130]

[0131] in, The fuzzy set M corresponding to temperature T The membership function, L T is the highest temperature in the low temperature range or the lowest temperature in the medium temperature range, M T is the middle value of the medium temperature range, H T It is the lowest temperature value in the high temperature range or the highest temperature value in the medium temperature range.

[0132]

[0133] in, The fuzzy set H corresponding to temperature T The membership function of .

[0134] The fuzzy set membership functions of pressure (P), flow (F), and current (I) adopt the membership function similar to the above-mentioned temperature.

[0135] Temperature (T), pressure (P), flow (F), and current (I) are extracted through 19 sensors to obtain 19 feature vectors (including 8 temperature data, 8 pressure data, 2 flow data, and 1 current data), and each feature vector is associated with three fuzzy membership functions, ultimately obtaining 57 fuzzy extracted feature vectors.

[0136] Through the fuzzification process, the actual value of each input variable is converted into a fuzzy membership. The fuzzy membership of each variable is concatenated as a feature to obtain the internal feature vector:

[0137] F E =[T1,T2,…T8,P1,P2,…P8,F1,F2,I]

[0138] in,

[0139] The sound data and vibration data are input into fuzzy module 2 and fuzzy module 3 respectively to obtain the extracted feature vector, vibration feature F V and sound characteristics F S , and then perform multimodal data fusion.

[0140] (2) Multimodal data cross-fusion methods, such as Figure 3 shown.

[0141] (1) Cross-fusion between modes. Since the cross-fusion calculations under internal, vibration, and sound modes are the same, we take the internal and vibration feature fusion as an example and input the internal feature F E and vibration characteristics F V , based on the multi-head attention mechanism, linearly map the internal features and vibration features to generate multi-dimensional attention weights. Use the attention weights to capture the connection between the internal and vibration modal information and obtain the interaction features between the modalities. Output the first fusion feature F of the internal and vibration data EV .

[0142] (2) The first fusion feature F of the internal and vibration data is obtained EV , the second fusion feature F of vibration and sound data VS , the third fusion feature F of internal and sound data ES For the first fusion feature F EV and the third fusion feature F ES Both of them fuse the internal data. In order to balance the weight of the modality, the first fusion feature F is calculated separately. EV With the second fusion feature F VS , the third fusion feature F ES With the second fusion feature F VS The cosine similarity α and β ofEV and the third fusion feature F ES The weight of the modality.

[0143] (3) The first fusion feature F with weight EV , the third fusion feature F ES With the second fusion feature F VS After further fusion, the final multimodal fusion feature F is obtained as:

[0144] F = LayerNorm(αF EV +βF ES +F VS )

[0145] Where α = cos(sim(F EV ,F VS )), β=cos(sim(F ES ,F VS )). The multimodal fusion feature F is input into the fault diagnosis model for model training.

[0146] Module 3: Fault diagnosis model building module

[0147] This module builds a fault diagnosis model to detect the operating status of the air-conditioning chiller. The multimodal fusion feature F is input into the Fast-RCNN network model for training. During the training process, the model is optimized and the parameters are adjusted multiple times to build a fault diagnosis model with accurate fault diagnosis results. The specific steps of this module are as follows:

[0148] (1) The multimodal fusion feature F is divided into the training set F train and the test set F test , the training set F train Input into the fault diagnosis model (Fast-RCNN network model) for training.

[0149] (2) During the model training process, the cross entropy loss function and Adam optimization algorithm are used to optimize the model and adjust the parameters. The cross entropy loss function is calculated as:

[0150]

[0151] Among them, y is the true label, is the model prediction value, and n is the number of samples.

[0152] (3) After model training, input the test set F test Multimodal fusion data, the k-fold cross validation method is used to verify and evaluate the model. According to the evaluation results, the model structure and hyperparameters are further adjusted to obtain a fault diagnosis model with high accuracy.

[0153] Module 4: Real-time diagnostic module

[0154] The real-time diagnosis module uses the trained fault diagnosis model to evaluate the operating status of the air-conditioning chiller.

[0155] First, unlabeled real-time data is collected through the data acquisition and annotation module, and then the unlabeled real-time data is input into the real-time diagnosis module. Then, the multimodal data feature extraction and fusion module is called to perform fuzzy feature extraction and multimodal fusion on the internal data, vibration data and sound data collected in real time to obtain multimodal fusion data. Finally, the multimodal fusion data is input into the fault diagnosis model in the real-time diagnosis module to perform real-time fault diagnosis and obtain the diagnosis result.

[0156] The real-time diagnosis module saves the diagnosis results as text data and uploads them to the database. If the system finds a fault, it will issue an alarm, and the operation and maintenance personnel will repair it and upload the repair records to the database; if no fault is detected, the detection results will be uploaded to the database as regular maintenance data for later reference.

[0157] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An air-conditioning chiller fault diagnosis method based on multimodal data fusion, characterized in that: include: Acquire internal data inside the air-conditioning chiller, sound data outside the air-conditioning chiller, and vibration data outside the compressor; the internal data includes temperature data, pressure data, flow data, and current data; Perform fuzzy processing on each variable in the internal data, sound data and vibration data to obtain internal features, sound features and vibration features; The internal features, sound features and vibration features are cross-fused in pairs to obtain the first fusion feature, the second fusion feature and the third fusion feature; Calculate the first similarity between the first fusion feature and the second fusion feature, and calculate the second similarity between the second fusion feature and the third fusion feature, use the first similarity as the weight of the first fusion feature, and use the second similarity as the weight of the third fusion feature; The first fusion feature, the third fusion feature and the second fusion feature with weights are fused to obtain a multimodal fusion feature; Based on multimodal fusion features and the fault diagnosis model, the fault diagnosis results of the air-conditioning chiller are obtained.

2. The air-conditioning chiller fault diagnosis method based on multimodal data fusion according to claim 1 is characterized in that: The internal features, the sound features and the vibration features are cross-fused in pairs to obtain a first fusion feature, a second fusion feature and a third fusion feature, the method comprising: Based on the internal features and vibration features, a multi-head attention mechanism is used to linearly map the internal features and vibration features to generate a first multi-dimensional attention weight; the first multi-dimensional attention weight is used to obtain the interaction features between the modalities, and the first fusion features of the internal data and the vibration data are output; Based on the sound features and vibration features, a multi-head attention mechanism is used to linearly map the sound features and vibration features to generate a second multi-dimensional attention weight; the second multi-dimensional attention weight is used to obtain the interaction features between the modalities, and the second fusion features of the sound data and the vibration data are output; Based on internal features and sound features, a multi-head attention mechanism is used to perform linear mapping on internal features and sound features to generate third-dimensional attention weights. The third-dimensional attention weights are used to obtain the interaction features between modalities and output the third fusion features of internal data and sound data.

3. The air-conditioning chiller fault diagnosis method based on multimodal data fusion according to claim 1 is characterized in that: Each variable in the internal data is fuzzy processed to obtain internal characteristics. The method includes: defining fuzzy sets and membership functions for temperature data, pressure data, flow data and current data respectively, and associating temperature data, pressure data, flow data and current data to different fuzzy sets through membership functions; wherein one membership function corresponds to one fuzzy data set, and low-temperature membership function, medium-temperature membership function, high-temperature membership function, low-pressure membership function, medium-pressure membership function, high-pressure membership function, low-flow membership function, medium-flow membership function, high-flow membership function, low-current membership function, medium-current membership function and high-current membership function correspond to different fuzzy sets respectively.

4. The air-conditioning chiller fault diagnosis method based on multimodal data fusion according to claim 3 is characterized in that: The low temperature membership function is expressed by the following formula: in, The fuzzy set L corresponding to the temperature T The membership function, T min is the minimum temperature value in the low temperature range, L T It is the highest temperature value in the low temperature range or the lowest temperature value in the medium temperature range.

5. The air-conditioning chiller fault diagnosis method based on multimodal data fusion according to claim 3 is characterized in that: The medium temperature membership function is expressed by the following formula: in, The second fuzzy set M corresponding to temperature T The membership function, L T is the highest temperature in the low temperature range or the lowest temperature in the medium temperature range, M T is the middle value of the medium temperature range, H T It is the lowest temperature value in the high temperature range or the highest temperature value in the medium temperature range.

6. The air-conditioning chiller fault diagnosis method based on multimodal data fusion according to claim 3 is characterized in that: The high temperature membership function is expressed by the following formula: in, The fuzzy set H corresponding to temperature T The membership function, M T is the middle value of the medium temperature range, H T is the lowest temperature in the high temperature range or the highest temperature in the medium temperature range, T max It is the highest temperature value in the high temperature range.

7. The air-conditioning chiller fault diagnosis method based on multimodal data fusion according to claim 3 is characterized in that: Each variable in the internal data is fuzzified to obtain internal features. The methods include: The actual value of each variable in the fuzzy set is converted into fuzzy membership, and the fuzzy membership of each variable is concatenated as a feature to obtain the internal feature.

8. The air-conditioning chiller fault diagnosis method based on multimodal data fusion according to claim 3 is characterized in that: Each variable in the sound data and the vibration data is fuzzy processed to obtain the sound features and the vibration features, and the method includes: Define fuzzy sets and membership functions for sound data and vibration data respectively, and associate the sound data and vibration data to different fuzzy sets through the membership functions; The actual value of each variable in the fuzzy set corresponding to the sound data is converted into a fuzzy membership degree, and the fuzzy membership degree of each variable is used as a feature for splicing to obtain the sound feature; The actual value of the fuzzy set corresponding to the vibration data is converted into the fuzzy membership, which is the vibration feature.

9. The air-conditioning chiller fault diagnosis method based on multimodal data fusion according to claim 1, characterized in that: The temperature data include: the refrigerant temperature at the condenser inlet, the refrigerant temperature at the condenser outlet, the refrigerant temperature at the evaporator inlet, the refrigerant temperature at the evaporator outlet, the refrigerant temperature at the compressor suction pipe, the refrigerant temperature at the compressor exhaust pipe, the refrigerant temperature at the front end of the throttling device, and the refrigerant temperature at the rear end of the throttling device; The pressure data include: the refrigerant pressure at the condenser inlet, the refrigerant pressure at the condenser outlet, the refrigerant pressure at the evaporator inlet, the refrigerant pressure at the evaporator outlet, the refrigerant pressure at the compressor suction pipe, the refrigerant pressure at the compressor exhaust pipe, the refrigerant pressure at the front end of the throttling device, and the refrigerant pressure at the rear end of the throttling device; The flow data includes: the refrigerant flow at the condenser inlet and the refrigerant flow at the evaporator inlet; The current data includes: the working current of the compressor; The sound data includes abnormal noise data outside the condenser, abnormal noise data outside the evaporator, abnormal noise data outside the compressor, and abnormal noise data outside the throttling device.

10. Air conditioning chiller fault diagnosis system based on multimodal data fusion, characterized in that: include: A temperature sensor installed inside the air-conditioning chiller, a pressure sensor installed outside the air-conditioning chiller, a flow sensor installed outside the air-conditioning chiller, a current sensor installed in the compressor power line, a condenser microphone installed outside the air-conditioning chiller, a condenser accelerometer installed on the compressor housing, and a processor; the processor is connected to the temperature sensor, the pressure sensor, the flow sensor, the current sensor, the condenser microphone, and the condenser accelerometer; A temperature sensor is used to collect temperature data, a pressure sensor is used to collect pressure data, a flow sensor is used to collect flow data, a current sensor is used to collect current data, a capacitor microphone is used to collect sound data, and a capacitive accelerometer is used to collect vibration data; Internal data include: temperature data, pressure data, flow data and current data; The processor is used to perform fuzzy processing on each variable in the internal data, sound data and vibration data to obtain internal features, sound features and vibration features; cross-fuse the internal features, sound features and vibration features in pairs to obtain the first fusion feature, the second fusion feature and the third fusion feature; respectively calculate the first similarity of the first fusion feature and the second fusion feature, and calculate the second similarity of the second fusion feature and the third fusion feature, use the first similarity as the weight of the first fusion feature, and use the second similarity as the weight of the third fusion feature; fuse the first fusion feature, the third fusion feature and the second fusion feature with weights to obtain multimodal fusion features; based on the multimodal fusion features, adopt a fault diagnosis model to obtain the fault diagnosis result of the air-conditioning chiller.

Images