A method and system for identifying faults of intelligent fire fighting equipment

Abstract

The present application provides a method and system for intelligent fire equipment fault identification, which relates to the technical field of fire equipment fault identification. The method comprises the steps of data collection, model construction, model training and fault identification; the system comprises a data collection module, a model construction module, a model training module and a fault identification module. The present application constructs a fire equipment fault identification model to overcome the interference of mechanical characteristic data of fire equipment in a vibration or impact environment, identify the real equipment fault signal, reduce the occurrence of misjudgment of faults, and improve the accuracy and reliability of fire equipment fault identification in a vibration or impact environment.

Description

Technical Field

[0001] The present application relates to the technical field of fire equipment fault identification, and in particular to a method and system for intelligent fire equipment fault identification. Background Art

[0002] As a core component of the fire safety system in modern society, the main purpose of fire equipment is to prevent and respond to various fire accidents, so as to protect the safety of life and property of personnel and the stable operation of public facilities. Fault detection of fire equipment is the key to ensuring fire safety. It can detect potential faults of fire equipment in time to ensure that the equipment can operate normally and play its role at critical moments. During actual operation, fire equipment will encounter many unfavorable factors such as vibration, temperature changes, electromagnetic interference, and dust and dirt accumulation. These unfavorable factors will cause the fault signal to be weak, the fault characteristics to be vague, and the fault type to be difficult to distinguish, which greatly increases the difficulty of identifying fire equipment faults.

[0003] The Chinese invention patent with the application publication date of October 3, 2023 and application publication number CN116843321A provides a method, system, equipment and storage medium for identifying fire equipment faults. It collects the status signals of fire equipment, analyzes the fault data and fault recovery data according to the preset fault filtering analysis algorithm, and determines the fire equipment fault.

[0004] The above technical solution only analyzes the failure of fire-fighting equipment based on the collected status signals of fire-fighting equipment. However, in some industrial manufacturing sites or transportation sites, the fire-fighting equipment will be intermittently in a vibration and shock environment due to the operation of the manufacturing equipment or transportation equipment. In such a vibration and shock environment, the operating state of the fire-fighting equipment is easily affected by mechanical dynamic characteristics such as vibration or shock, thereby interfering with the data collection and analysis of the fire-fighting equipment, and further causing false alarms or missed alarms in the fault detection results, reducing the accuracy and reliability of the fire-fighting equipment failure prediction. Summary of the invention

[0005] In order to improve the accuracy of fire equipment failure prediction, the present application provides a smart fire equipment failure identification method and system.

[0006] In the first aspect, the present application provides a method for identifying faults of intelligent fire fighting equipment, which adopts the following technical solutions:

[0007] A method for identifying faults of intelligent fire fighting equipment, the method comprising:

[0008] Data collection: including real-time data collection and historical data collection;

[0009] Collect real-time data: collect real-time data of fire-fighting equipment;

[0010] Collect historical data: collect historical data of fire-fighting equipment and fault type labels corresponding to the historical data, wherein the historical data includes historical equipment operation data and historical mechanical characteristic data;

[0011] Model construction: constructing a fire equipment fault identification model, wherein the fire equipment fault identification model includes a hidden Markov sub-model and a fault classification sub-model;

[0012] Model training: including first classification, first processing and second classification;

[0013] First classification: taking the historical data of the fire-fighting equipment and the fault type labels corresponding to the historical data as the input of the hidden Markov sub-model to obtain the first data;

[0014] First processing: splicing the first data with the historical data of the fire-fighting equipment and recording them as the second data;

[0015] Second classification: taking the second data as the input of the fault classification sub-model and outputting the fault type label corresponding to the second data;

[0016] Fault self-checking: Input the real-time data of fire-fighting equipment into the trained fire-fighting equipment fault identification model, and output the fault type label of the real-time data.

[0017] By adopting the above technical solution, the mechanical characteristic data of fire-fighting equipment is considered in a vibration and shock environment, combined with the equipment operation data of the fire-fighting equipment, and the hidden Markov sub-model is used to capture the hidden state corresponding to the change of the mechanical characteristic data, and the law related to the equipment failure is excavated. Then, the fault classification sub-model is used to determine the fault type of the fire-fighting equipment, which helps to overcome the interference of the vibration and shock environment state on the equipment fault analysis, more accurately identify the real equipment failure, reduce the occurrence of misjudgment of faults, and improve the accuracy and reliability of fire-fighting equipment fault identification in the vibration and shock environment state.

[0018] Optionally, after executing the step of model building and before executing the step of first classification, the method further includes:

[0019] Time threshold setting: Different equipment usage time labels are set according to the service life of fire-fighting equipment;

[0020] Fault type label classification: Different fault types are classified according to the fault classification labels of historical data;

[0021] Define implicit states: Based on the collected historical data, label the historical data with the corresponding equipment usage time tags, match the equipment usage time tags of the historical data with the corresponding fault types, and use the matched data as the implicit states of the hidden Markov sub-model;

[0022] First iteration training: input historical data into the hidden Markov sub-model for probability calculation to obtain the forward probability and backward probability of the hidden state, recorded as the first probability, update the observation probability matrix according to the first probability, optimize the hidden Markov sub-model, and obtain the optimized hidden Markov sub-model, recorded as the new hidden Markov sub-model.

[0023] By adopting the above technical solution, considering the differences in the entire life cycle of fire-fighting equipment, matching the historical usage time of the current historical data with the corresponding fault type as the implicit state of the hidden Markov sub-model, it helps the model to more accurately judge whether the current data changes are caused by the normal aging stage characteristics of the equipment, or by abnormal faults or simply by the impact of the vibration shock environment, reducing the possibility of misjudgment of faults due to equipment aging and environmental influences, and improving the accuracy of fault analysis.

[0024] Optionally, after the step of setting the execution time threshold and before the step of classifying the fault type labels, the method further includes:

[0025] First judgment: determine whether the historical equipment operation data in the current historical data are all within the preset operation threshold range:

[0026] If yes, the types of historical equipment operation data in the current historical data are all marked as normal operation data, and the second determination step is performed;

[0027] If not, the type of historical equipment operation data that exceeds the preset operation threshold range is marked as abnormal operation data, and the type of historical equipment operation data that does not exceed the preset operation threshold range is marked as normal operation data, and the second determination step is performed;

[0028] Second judgment: judge whether the historical mechanical characteristic data in the current historical data is within the preset characteristic threshold range:

[0029] If so, the types of historical mechanical characteristic data in the current historical data are all marked as normal characteristic data;

[0030] If not, the type of historical mechanical characteristic data that exceeds the preset characteristic threshold range is marked as abnormal characteristic data, and the type of historical mechanical characteristic data that does not exceed the preset characteristic threshold range is marked as normal characteristic data;

[0031] First identification: Match the type of the annotated historical mechanical characteristic data and the type of the historical equipment operation data in the current historical data with the corresponding fault type label, record it as the first matching label, and use the first matching label as the new fault type label of the current historical data.

[0032] By adopting the above technical solution, the impact of equipment operating status and mechanical characteristic changes on fault types is comprehensively considered, and the equipment operating data and mechanical characteristic data are labeled and matched with the fault type. This improves the correlation between the equipment operating data and mechanical characteristic data and the fault type, and helps the model understand the combination and change patterns of equipment operating data and mechanical characteristic data in complex situations based on more accurate fault type labels, so as to accurately judge the equipment status and reduce the misjudgment of faults due to the vibration and shock environment affecting the mechanical characteristics of fire-fighting equipment.

[0033] Optionally, after executing the step of building the model and before executing the step of setting the time threshold, the method further includes:

[0034] Fault digitization: Perform one-hot encoding on the fault classification label to obtain the first label;

[0035] First screening: calculating the correlation between the historical equipment operation data and the first tag, recording it as the first correlation data, screening out the historical equipment operation data whose first correlation data is higher than the preset correlation state threshold value as the new historical equipment operation data;

[0036] Second screening: Calculate the correlation between the historical mechanical feature data and the first label, record it as the second correlation data, and screen out the historical mechanical feature data whose second correlation data is higher than the preset correlation feature threshold as the new historical mechanical feature data.

[0037] By adopting the above technical solution, equipment operation data and mechanical characteristic data that are not strongly correlated with the fault type are screened out, thereby reducing the impact of some irrelevant state data changes caused by the vibration and shock environment on fault judgment, which helps to reduce misjudgment of faults caused by environmental influences.

[0038] Optionally, after performing the first classification step and before performing the first processing step, the method further includes:

[0039] Probability difference calculation: Calculate the probability difference of the same data under different implicit states in the first data, record it as the first probability difference, and use the first probability difference and the first data as the new first data.

[0040] By adopting the above technical solution, introducing the probability difference, highlighting the data changes caused by equipment failure under the normal influence of the vibration shock environment, it is helpful to effectively distinguish between the fault situation and the normal environmental influence, and help reduce the misjudgment of faults. At the same time, using the first probability difference and the first data as the new first data helps the model better adapt to the complex data change pattern under the vibration shock environment, improves the accuracy of fault judgment, and helps reduce the misjudgment of faults caused by the influence of the vibration shock environment.

[0041] Optionally, the fault classification submodel includes a decision tree layer and a neural network layer, and after performing the first processing step and before performing the second classification step, further includes:

[0042] Define the root node: calculate the information gain ratio of all data features in the second data, and select the data feature with the maximum information gain ratio as the root node;

[0043] First fault classification training: input the second data into the decision tree layer for inference, calculate the residual of the predicted fault type label and the fault type label corresponding to the historical data, build a decision tree based on the calculated residual and the root node, perform model inference, and output the fault prediction probability;

[0044] Second fault classification training: concatenate the fault prediction probability with the second data to obtain the third data, use the third data as the sample training set, input the neural network layer, perform model inference, define the feature loss function, and iteratively optimize the neural network layer by minimizing the feature loss function to obtain the trained fault classification sub-model as the new fault classification sub-model.

[0045] By adopting the above technical solution, the decision tree layer can continuously adjust the residual to update the internal judgment rules, which helps to accurately identify data fluctuations and fault signals under the influence of normal environment. In addition, the neural network layer is used to explore the distinction between faults and normal states in the vibration impact environment from a more comprehensive perspective. Combining the decision tree layer and the neural network layer can classify faults more comprehensively and accurately, which helps to reduce the misjudgment of faults due to the influence of the vibration impact environment, and improves the reliability and accuracy of the fault classification sub-model in judging fire equipment faults in vibration impact scenarios.

[0046] Optionally, the feature loss function is:

[0047]

[0048] Where n is the number of samples in the sample training set, Indicates the third data The weight of the data features, Indicates The first data features, Indicates The first The predicted value of a data feature, represents the balance coefficient, To indicate the The fault type label of samples is the true label, The output of the neural network layer The predicted fault type labels for samples.

[0049] By adopting the above technical solution, the key features that are strongly correlated with faults will be given more important weights using the feature loss function, so that the neural network layer can allocate learning focus according to the importance of each feature to accurately judge the fault during training, and focus on learning the performance laws of key features under the influence of different environments, so as to achieve the technical effect of more accurately grasping the key factors of fault judgment, which helps to reduce the misjudgment of faults caused by failure to accurately capture the changes in key features.

[0050] Optionally, after executing the step of the first processing and before executing the step of defining the root node, the method further includes:

[0051] Third screening: calculating the correlation between the data features in the second data, recorded as third correlation data, and screening out the data feature group whose third correlation data is greater than a preset screening threshold, recorded as the first data group;

[0052] Feature selection: Use principal component analysis to convert the two data features of each data group in the first data group into fused data features, recorded as fused data features, collect the fused data features of each data group to obtain a fused data feature group, and use the fused data feature group to replace the data of the first data group in the second data to obtain the replaced data as the new second data.

[0053] By adopting the above technical solution, the principal component analysis method is used to fuse the data features in the first data group, thereby reducing the dimension of the data features while retaining the main information in the data features, so that the model can grasp the core features of the data more efficiently when processing the data features, which helps the model to more accurately distinguish between data changes under normal environmental influences and abnormal changes when a fault occurs in a vibration and impact environment, and reduce the misjudgment of faults caused by complex and difficult to distinguish data.

[0054] In a second aspect, the present application provides a smart fire equipment fault identification system, wherein the segmentation system is applicable to the segmentation method described in any one of the first aspects above, and the segmentation system includes: adopting the following technical solution:

[0055] A smart fire equipment fault identification system, the system comprising:

[0056] Data collection module, including real-time data collection module and historical data collection module;

[0057] Real-time data collection module, used to collect real-time data of fire-fighting equipment;

[0058] A historical data collection module is used to collect historical data of fire-fighting equipment and fault type labels corresponding to the historical data, wherein the historical data includes historical equipment operation data and historical mechanical characteristic data;

[0059] A model building module, used to build a fire equipment fault identification model, wherein the fire equipment fault identification model includes a hidden Markov sub-model and a fault classification sub-model;

[0060] The model training module is connected to the data acquisition module and the model building module in communication, and includes a first classification unit, a first processing unit, and a second classification unit;

[0061] A first classification unit is used to use the historical data of the fire-fighting equipment and the fault type label corresponding to the historical data as the input of the hidden Markov sub-model to obtain first data;

[0062] A first processing unit is used to combine the first data with the historical data of the fire-fighting equipment to record the data as second data;

[0063] A second classification unit, used to use the second data as an input of a fault classification sub-model and output a fault type label corresponding to the historical data;

[0064] The fault identification unit is connected to the data acquisition module and the model building module for inputting the real-time data of the fire-fighting equipment into the trained fire-fighting equipment fault identification model and outputting the fault type label of the real-time data.

[0065] By adopting the above technical solution, the mechanical characteristic data of the fire-fighting equipment is considered under the vibration and shock environment. The data acquisition module collects the mechanical characteristic data and equipment operation data of the fire-fighting equipment. The fire-fighting equipment fault identification model constructed by the model construction module is used to identify the fault type of the fire-fighting equipment. This overcomes the interference of the vibration and shock environment state on the equipment fault analysis, can more accurately identify the real equipment failure, reduce the occurrence of misjudgment of faults, and improve the accuracy and reliability of fire-fighting equipment fault identification under the vibration and shock environment state.

[0066] In summary, the present application includes at least one of the following beneficial technical effects:

[0067] 1. Considering the mechanical characteristic data of fire-fighting equipment in a vibration and shock environment, combined with the equipment operation data of the fire-fighting equipment, the hidden Markov sub-model is used to capture the hidden state corresponding to the change of mechanical characteristic data, and the law related to equipment failure is excavated. Then, the fault classification sub-model is used to determine the fault type of the fire-fighting equipment, which helps to overcome the interference of the vibration and shock environment state on the equipment fault analysis, more accurately identify the real equipment failure, reduce the occurrence of misjudgment of faults, and improve the accuracy and reliability of fire-fighting equipment fault identification in a vibration and shock environment.

[0068] 2. Considering the differences in the entire life cycle of fire-fighting equipment, the historical usage time of the current historical data is matched with the corresponding fault type as the implicit state of the hidden Markov model. This helps the model to more accurately judge whether the current data changes are caused by the normal aging characteristics of the equipment, or by abnormal faults or simply by the impact of the vibration shock environment. This reduces the possibility of misjudgment of faults due to equipment aging and environmental influences, and improves the accuracy of fault analysis.

[0069] 3. Comprehensively consider the impact of equipment operating status and mechanical characteristic changes on fault types, divide the labels of equipment operating data and mechanical characteristic data, and match the divided equipment operating data and mechanical characteristic data with the fault type, which improves the correlation between equipment operating data and mechanical characteristic data and fault types. It helps the model to understand the combination and change patterns of equipment operating data and mechanical characteristic data in complex situations based on more accurate fault type labels, so as to accurately judge the equipment status and reduce the misjudgment of faults due to the impact of vibration and shock environment on the mechanical characteristics of fire-fighting equipment.

[0070] 4. Combining the decision tree layer and the neural network layer can classify faults more comprehensively and accurately, which helps to reduce the misjudgment of faults due to the influence of vibration and shock environment, and improves the reliability and accuracy of the fault classification sub-model in judging fire equipment faults under vibration and shock scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0071] Figure 1 is a flow chart of Example 1 of the present application;

[0072] Figure 2 This is a flowchart of the S3 model training of Example 1 of the present application;

[0073] Figure 3 This is a flowchart of the first model training in S31 of Example 1 of the present application;

[0074] Figure 4 This is a flowchart of the implicit state definition in S312 of Example 1 of the present application;

[0075] Figure 5 This is a flowchart of the second model training in S34 of Example 1 of the present application. DETAILED DESCRIPTION

[0076] The following combination Figures 1 to 5 This application is described in further detail.

[0077] Embodiment 1: This embodiment discloses a method for identifying faults of intelligent fire fighting equipment, such as Figure 1As shown, the method includes: collecting real-time data of fire-fighting equipment, historical data of fire-fighting equipment, and fault type labels corresponding to the historical data, building a fire-fighting equipment fault identification model, the fire-fighting equipment fault identification model includes a hidden Markov sub-model and a fault classification sub-model, inputting the historical data of the fire-fighting equipment and the fault type labels corresponding to the historical data into the fire-fighting equipment fault identification model for model training, obtaining a trained fire-fighting equipment fault identification model, inputting the real-time data of the fire-fighting equipment into the trained fire-fighting equipment fault identification model, and outputting the fault type labels of the real-time data. This embodiment includes the following steps:

[0078] S1 data collection: includes S11 real-time data collection and S12 historical data collection.

[0079] S11 collects real-time data: collects real-time data of fire-fighting equipment, including equipment operation data and mechanical characteristic data. Equipment operation data includes: equipment start and stop status, working mode, equipment temperature, pressure, equipment electrical characteristics (including current, voltage and power), water flow, ambient temperature, ambient humidity and ambient air pressure. Mechanical characteristic data includes equipment vibration characteristics (including vibration acceleration, vibration velocity, vibration displacement signal and vibration spectrum characteristics), equipment impact characteristics (including impact force, impact force duration, transient impact signal, impact power spectrum density), equipment stress data, equipment speed and equipment torque.

[0080] In this embodiment, the equipment start-stop state includes the start-up and shutdown of the equipment, and the working mode includes operation, standby and maintenance. The equipment start-stop state and working mode can be obtained by direct observation and recording. The equipment temperature can be collected by a temperature sensor installed on the parts of the fire-fighting equipment, and the ambient temperature can be collected by a temperature sensor installed in the machine room, the inside of the box or the external environment where the fire-fighting equipment is located. The pressure is collected by a pressure sensor installed in the pump outlet of the fire-fighting equipment, the key node of the hydraulic system or the pipeline system. The electrical characteristics of the fire-fighting equipment are collected by electrical sensors such as current sensors and voltage sensors installed on the power distribution cabinet or the motor input end of the fire-fighting equipment. The power can be calculated by the collected current and voltage. The water flow is collected by a flow sensor installed in the water outlet or pipeline of the fire-fighting pump. The ambient humidity is collected by a humidity sensor installed in the environment where the fire-fighting equipment is located. The ambient air pressure is collected by an air pressure sensor installed in the environment where the fire-fighting equipment is located. The vibration acceleration, vibration velocity and vibration displacement signals in the vibration characteristics of the equipment are collected by a vibration sensor (including an acceleration sensor, a velocity sensor or a displacement sensor) installed on the key mechanical parts of the fire-fighting equipment (such as the motor housing, the bearing seat, the pump housing, etc.). The vibration spectrum characteristics are obtained by analyzing the collected vibration acceleration, vibration velocity and vibration displacement signals through a spectrum analyzer. The impact force and impact force duration in the equipment impact characteristics are collected through impact sensors installed on the fire-fighting equipment frame or parts susceptible to impact (such as the shell). The transient impact signal in the equipment impact characteristics is collected through strain sensors and oscilloscopes, and the impact power spectrum density is obtained through corresponding signal processing. Equipment stress data is collected through strain gauge sensors installed on key load-bearing parts of fire-fighting equipment (such as bearing outer rings, connecting parts or supporting structures). The equipment speed is collected through speed sensors installed on rotating parts of fire-fighting equipment (such as motors or pump shafts). The equipment torque is collected through torque sensors installed on the transmission shaft or coupling of fire-fighting equipment.

[0081] In this embodiment, when collecting the real-time data of the fire-fighting equipment, the equipment operation data and mechanical characteristic data of the fire-fighting equipment collected at the same time point are taken as a set of real-time data of the fire-fighting equipment.

[0082] S12 collects historical data: collects historical data of fire-fighting equipment and the fault type labels corresponding to the historical data. The historical data of fire-fighting equipment includes historical equipment operation data and historical mechanical characteristic data. The historical equipment operation data includes: historical equipment start and stop status, historical working mode, historical service life, historical equipment temperature, historical pressure, historical equipment electrical characteristics (including historical current, historical voltage and historical power), historical water flow, historical ambient temperature, historical ambient humidity and historical ambient air pressure. The historical mechanical characteristic data includes historical equipment vibration characteristics (including historical vibration acceleration, historical vibration velocity, historical vibration displacement signal and historical vibration spectrum characteristics), historical equipment impact characteristics (including historical impact force, historical impact force duration, historical transient impact signal, historical impact power spectrum density), historical equipment stress data, historical equipment speed and historical equipment torque. The fault type labels include normal state, mechanical fault, electrical fault, abnormal temperature fault and abnormal flow fault. Mechanical faults include mild mechanical fault, moderate mechanical fault and severe mechanical fault. Electrical faults include mild electrical fault, moderate electrical fault and severe electrical fault. The temperature abnormality fault includes a slight temperature abnormality fault, a moderate temperature abnormality fault and a severe temperature abnormality fault. The flow abnormality fault includes a slight flow abnormality fault, a moderate flow abnormality fault and a severe flow abnormality fault.

[0083] In this embodiment, the historical equipment operation data and historical mechanical characteristic data collected at the same time point and their corresponding fault type labels are used as a group of historical data of fire fighting equipment.

[0084] S2 Model Construction: Construct a fire equipment fault identification model, which includes a hidden Markov sub-model and a fault classification sub-model. The fault classification sub-model includes a decision tree layer and a neural network layer.

[0085] In this embodiment, a continuous hidden Markov model (Continuous-HMM) is used as the basic model of the hidden Markov sub-model. The decision tree layer of the fault classification sub-model is constructed based on the C4.5 decision tree algorithm, and the neural network layer of the fault classification sub-model is constructed using a multi-layer perceptron.

[0086] S3 model training: including S31 first model training, S32 first classification, S33 first processing, S34 second model training and S35 second classification, such as Figure 2 shown.

[0087] S31 first model training: including S311 data screening, S312 implicit state definition and S313 first iteration training, such as Figure 3 shown.

[0088] S311 data screening: including S3111 fault quantification, S3112 first screening and S3113 second screening.

[0089] S3111 Fault digitization: Perform one-hot encoding on the fault classification labels in the historical data, convert the fault classification labels into vector form, and obtain a matrix, which is recorded as the first label.

[0090] S3112 First screening: Use the Pearson correlation coefficient to calculate the correlation between each historical equipment operation data in the historical data and the first label, record it as the first correlation data, compare the first correlation data with the preset correlation state threshold, screen out the historical equipment operation data whose first correlation data is higher than the preset correlation state threshold, and use the screened historical equipment operation data as the new historical equipment operation data.

[0091] S3113 Second screening: Use the Pearson correlation coefficient to calculate the correlation between each historical mechanical feature data in the historical data and the first label, record it as the second correlation data, compare the second correlation data with the preset correlation feature thresholds, screen out the historical mechanical feature data whose second correlation data is higher than the preset correlation feature threshold, and use the screened historical mechanical feature data as new historical mechanical feature data.

[0092] In this embodiment, by screening out historical equipment operation data and historical mechanical characteristic data that are not strongly correlated with the fault type, the impact of some irrelevant state data changes caused by the vibration impact environment on fault judgment is reduced, which helps to reduce misjudgment of faults caused by environmental influences.

[0093] S312 Implicit state definition: including S3121 time threshold setting, S3122 first judgment, S3123 second judgment, S3124 first identification, S3125 fault type label classification and S3126 definition of implicit state, such as Figure 4 shown.

[0094] S3121 Time threshold setting: Determine the service life of the fire-fighting equipment according to the design of the fire-fighting equipment. According to the normal operation status change curve of the fire-fighting equipment, divide the time intervals of the service life of the fire-fighting equipment into three equipment use time labels: early use, mid-use and late use. Output the comparison table of equipment service life and equipment use time label.

[0095] S3122 First judgment: determine whether the historical equipment operation data in the current historical data are all within the preset operation threshold range.

[0096] If so, the types of historical equipment operation data in the current historical data are all marked as normal operation data, and the second determination in S3123 is performed.

[0097] If not, the type of historical equipment operation data that exceeds the preset operation threshold range is marked as abnormal operation data, and the type of historical equipment operation data that does not exceed the preset operation threshold range is marked as normal operation data, and the second judgment of S3123 is executed.

[0098] S3123 Second judgment: determine whether the historical mechanical characteristic data in the current historical data are all within the preset characteristic threshold range.

[0099] If so, the types of historical mechanical characteristic data in the historical data are all marked as normal characteristic data.

[0100] If not, the type of historical mechanical characteristic data that exceeds the preset characteristic threshold range is marked as abnormal characteristic data, and the type of historical mechanical characteristic data that does not exceed the preset characteristic threshold range is marked as normal characteristic data.

[0101] S3124 First identification: Match the type of the annotated historical mechanical characteristic data and the type of the historical equipment operation data in the historical data with the fault type label corresponding to the current historical data, record it as the first matching label, and use the first matching label as the new fault type label of the current historical data. For example, the historical equipment operation data of the current historical data do not exceed the preset operation threshold range and the equipment speed in the historical mechanical characteristic data exceeds the preset speed characteristic threshold range, and the fault type label of the current historical data is a slight mechanical fault, then the matched fault type label is briefly described as: the historical equipment operation data is normal operation data-the historical equipment speed is abnormal characteristic data-the other historical mechanical characteristic data except the historical equipment speed is normal operation data-mild mechanical fault. In this embodiment, it is explained in abbreviated form that when matching, the type of each historical equipment operation data and the type of each historical mechanical characteristic data need to be listed to match the fault type label, and then all the listed types of historical equipment operation data and historical mechanical characteristic data and the fault type label of the current historical data are used as the new fault type label of the current historical data.

[0102] S3125 Fault type label classification: Different fault types are classified according to the fault classification labels of historical data. For example, if the fault type label of historical data is historical equipment operation data is normal operation data-historical equipment speed abnormal characteristic data-other historical mechanical characteristic data except historical equipment speed is normal operation data-mild mechanical fault, then the fault types classified are historical equipment operation data is normal operation data-historical equipment speed abnormal characteristic data-other historical mechanical characteristic data except historical equipment speed is normal operation data-mild mechanical fault.

[0103] S3126 defines the implicit state: according to the historical service life in the historical data, the historical data is labeled with the equipment use time label according to the comparison table of equipment service life and equipment use time label, the equipment use time label of the current historical data is matched with the corresponding fault type, and the matched data is used as the implicit state of the hidden Markov sub-model. For example, the equipment use time label of the current historical data is the initial use, and the fault type is the historical equipment operation data is normal operation data-historical equipment speed abnormal feature data-other historical mechanical feature data except historical equipment speed is normal operation data-mild mechanical fault, then the matched data is the initial use-historical equipment operation data is normal operation data-historical equipment speed abnormal feature data-other historical mechanical feature data except historical equipment speed is normal operation data-mild mechanical fault.

[0104] In this embodiment, the differences in the entire life cycle of fire-fighting equipment are taken into account, the impact of equipment operating status and mechanical characteristic changes on fault types are considered, the equipment operating data and mechanical characteristic data are labeled, and the divided equipment operating data and mechanical characteristic data are matched with the fault type, and the historical usage time of the current historical data is matched with the corresponding fault type. This helps the model to understand the combination and change patterns of equipment operating data and mechanical characteristic data in complex situations based on more accurate fault type labels under the influence of normal aging characteristics of the equipment, so as to accurately judge the equipment status, reduce the possibility of misjudgment of faults due to equipment aging and environmental influences, and improve the accuracy of fault analysis.

[0105] S313 first iteration training: initialize the parameters of the hidden Markov sub-model, input historical data into the hidden Markov sub-model, use the forward algorithm to calculate the forward probability, use the backward algorithm to calculate the backward probability, record the forward probability and the backward probability as the first probability, update the observation probability matrix according to the first probability, optimize the hidden Markov sub-model, and obtain the optimized hidden Markov sub-model, which is recorded as the new hidden Markov sub-model.

[0106] S32 First classification: taking the historical data of the fire-fighting equipment and the fault type labels corresponding to the historical data as inputs of the hidden Markov sub-model to obtain first data.

[0107] S33 the first processing includes S331 probability difference calculation and S332 data splicing.

[0108] S331 Probability difference calculation: Calculate the probability difference of the same data under different implicit states in the first data, record it as the first probability difference, and use the first probability difference and the first data as the new first data.

[0109] S332 Data splicing: Extract the first probability difference of the same data type as the historical equipment operation data in the historical data from the first data, record it as the first source data, multiply and splice the first source data with the historical equipment operation data, and record the spliced ​​data as the first spliced ​​data. Extract the first probability difference of the same data type as the historical mechanical characteristic data in the historical data from the first data, record it as the second source data, multiply and splice the second source data with the historical mechanical characteristic data, and record the spliced ​​data as the second spliced ​​data. The first spliced ​​data, the second spliced ​​data, and the first data are recorded as the second data.

[0110] In this embodiment, the probability difference is introduced to highlight the data changes caused by equipment failure under the normal influence of the vibration shock environment, which helps to effectively distinguish between the fault situation and the normal environmental influence and helps to reduce the misjudgment of faults.

[0111] S34 second model training: including S341 second processing, S342 defining the root node, S343 first fault classification training and S344 second fault classification training, such as Figure 5 shown.

[0112] S341 second processing: includes S3411 third screening and S3412 feature selection.

[0113] S3411 Third screening: Use the Pearson correlation coefficient to calculate the correlation between the various data features in the second data, record it as the third correlation data, compare the third correlation data with the preset screening threshold, screen out the third correlation data greater than the preset screening threshold, take the two data features for calculating the third correlation data as a group of data, and record a group of data greater than the preset screening threshold as the first data group.

[0114] S3412 Feature selection: Using principal component analysis, the two data features of each data group in the first data group are converted into fused data features through linear transformation, which are used as the fused data features of each data group. The fused data features of each data group are collected and recorded as a fused data feature group. The fused data feature group is used to replace the data of the first data group in the second data to obtain the replaced data as the new second data.

[0115] The two data features of each data group in the first data group are standardized to obtain features A and B, and the covariance matrix of features A and B is calculated. The covariance matrix is ​​solved to obtain the eigenvalues ​​of features A and B, and the homogeneous linear equations are solved for each eigenvalue to obtain the eigenvector of each eigenvalue. The variance contribution rate of each eigenvalue is calculated, and then the cumulative variance contribution rate is calculated. When the cumulative variance contribution rate reaches the preset contribution rate threshold, the eigenvalues ​​of the cumulative variance contribution rate involved in the calculation are selected and recorded as the principal components of features A and B. Linearly combine features A and B with the eigenvectors corresponding to each principal component to obtain the principal component data features, which are recorded as the fusion data features of the current data group.

[0116] For example, the principal components of feature A and feature B are PC1 and PC2 respectively, and the eigenvector of PC1 is V1=(a 11 ,a 12 ), the eigenvector of PC2 is V2=(a 21 ,a 22 ), then the data characteristics of PC1 are PC1=a 11 ×Feature A+a 12 × Feature B, the data feature of PC2 is PC2=a 21 ×Feature A+a 22 ×Feature B.

[0117] S342 defines a root node: calculates the information gain of all data features in the second data, normalizes the information gain of each feature, obtains the information gain ratio of each data feature, and selects the data feature with the maximum information gain ratio as the root node.

[0118] S343 First fault classification training: input the second data into the decision tree layer for model inference, calculate the residual of the predicted fault type label and the fault type label corresponding to the historical data, build a decision tree based on the calculated residual and the root node, perform model inference, and output the fault prediction probability.

[0119] S344 Second fault classification training: directly concatenate the fault prediction probability with the second data, record the concatenated data as the third data, input the third data as the sample training set into the neural network layer, perform model inference, define the feature loss function, iteratively optimize the neural network layer by minimizing the feature loss function, and obtain the trained fault classification sub-model as the new fault classification sub-model.

[0120] The feature loss function is:

[0121]

[0122] Where n is the number of samples in the sample training set, Indicates the third data The weight of the data features, Indicates The first data features, Indicates The first The predicted value of a data feature, represents the balance coefficient, To indicate the The fault type label of samples is the true label, is the output of the neural network layer The predicted fault type labels for samples.

[0123] In this embodiment, a grid search method is used to select a series of candidate values, and evaluate the fault classification sub-models in different The performance under the value is selected to make the model performance index optimal The value is taken as the final balance coefficient.

[0124] In this embodiment, a self-attention layer is added to the neural network layer. When the third data is input into the neural network layer as a sample training set, the attention score of each data feature in the third data is calculated by the self-attention layer, and the calculated attention score is normalized to obtain the weight of each data feature. The sum of the weights of all different data features in the third data is 1.

[0125] S35 Second classification: Use the second data as the input of the fault classification sub-model, and output the fault type label corresponding to the second data.

[0126] S4 Fault Identification: Input the real-time data of fire-fighting equipment into the trained fire-fighting equipment fault identification model and output the fault type label of the real-time data.

[0127] In this embodiment, the mechanical characteristic data of the fire-fighting equipment is considered in a vibration and shock environment, combined with the equipment operation data of the fire-fighting equipment, and a constructed fire-fighting equipment fault identification model is used to help overcome the interference of the vibration and shock environment state on the equipment fault analysis, more accurately identify the real equipment failure, reduce the occurrence of misjudgment of faults, and improve the accuracy and reliability of fire-fighting equipment fault identification in the vibration and shock environment state.

[0128] Embodiment 2: This embodiment discloses a smart fire equipment fault identification system, the system comprising:

[0129] The data acquisition module includes a real-time data acquisition module and a historical data acquisition module.

[0130] The real-time data collection module is used to collect real-time data of fire-fighting equipment.

[0131] The historical data collection module is used to collect historical data of fire-fighting equipment and fault type labels corresponding to the historical data, wherein the historical data includes historical equipment operation data and historical mechanical characteristic data.

[0132] The model building module is used to build a fire equipment fault identification model, wherein the fire equipment fault identification model includes a hidden Markov sub-model and a fault classification sub-model.

[0133] The model training module is connected to the data acquisition module and the model building module in communication, and includes a first model training unit, a first classification unit, a first processing unit, a second model training unit and a second classification unit.

[0134] The first model training unit is used to input historical data into the hidden Markov sub-model for probability calculation, obtain the forward probability and backward probability of the hidden state, recorded as the first probability, update the observation probability matrix according to the first probability, optimize the hidden Markov sub-model, and obtain the optimized hidden Markov sub-model, recorded as the new hidden Markov sub-model.

[0135] The first classification unit is used to use the historical data of the fire-fighting equipment and the fault type labels corresponding to the historical data as inputs of the hidden Markov sub-model to obtain first data.

[0136] The first processing unit is communicatively connected with the first classification unit and the second model training unit, and includes a probability difference calculation subunit and a data splicing subunit.

[0137] The probability difference calculation subunit is used to calculate the probability difference of the same data under different implicit states in the first data, recorded as the first probability difference, and use the first probability difference and the first data as new first data.

[0138] The data splicing subunit is used to splice the first data with the historical data of the fire-fighting equipment and record them as second data.

[0139] The second model training unit is communicatively connected with the first processing unit and the second classification unit, and includes a root node definition subunit, a first fault classification training subunit, and a second fault classification training subunit.

[0140] A root node subunit is defined to calculate the information gain ratio of all data features in the second data, and the data feature with the maximum information gain ratio is selected as the root node.

[0141] The first fault classification training subunit is used to input the second data into the decision tree layer for reasoning, calculate the residual of the predicted fault type label and the fault type label corresponding to the historical data, build a decision tree based on the calculated residual and the root node, perform model reasoning, and output the fault prediction probability.

[0142] The second fault classification training subunit is used to concatenate the fault prediction probability with the second data to obtain third data, use the third data as a sample training set, input the third data into the neural network layer, perform model inference, define a feature loss function, and iteratively optimize the neural network layer by minimizing the feature loss function to obtain a trained fault classification sub-model as a new fault classification sub-model.

[0143] The second classification unit is used to use the second data as an input of the fault classification sub-model and output a fault type label corresponding to the historical data.

[0144] The fault identification unit is connected to the data acquisition module and the model building module for inputting the real-time data of the fire-fighting equipment into the trained fire-fighting equipment fault identification model and outputting the fault type label of the real-time data.

[0145] In this embodiment, the mechanical characteristic data of the fire-fighting equipment is considered in the vibration and shock environment, and the fire-fighting equipment fault identification model constructed by the model building module is used to identify the fault type of the fire-fighting equipment, thereby overcoming the interference of the vibration and shock environment state on the equipment fault analysis, being able to more accurately identify the real equipment failure, reducing the occurrence of misjudgment of faults, and improving the accuracy and reliability of fire-fighting equipment fault identification in the vibration and shock environment state.

[0146] The above are all preferred embodiments of the present application, and the protection scope of the present application is not limited thereto. Therefore, any equivalent changes made according to the structure, shape, and principle of the present application should be included in the protection scope of the present application.

Claims

1. A method for identifying faults of intelligent fire fighting equipment, characterized in that: The method comprises: Data collection: including real-time data collection and historical data collection; Collect real-time data: collect real-time data of fire-fighting equipment; Collect historical data: collect historical data of fire-fighting equipment and fault type labels corresponding to the historical data, wherein the historical data includes historical equipment operation data and historical mechanical characteristic data; Model construction: constructing a fire equipment fault identification model, wherein the fire equipment fault identification model includes a hidden Markov sub-model and a fault classification sub-model; the fault classification sub-model includes a decision tree layer and a neural network layer; Model training: including first classification, first processing, defining root node, first fault classification training, second fault classification training and second classification; First classification: taking the historical data of the fire-fighting equipment and the fault type labels corresponding to the historical data as the input of the hidden Markov sub-model to obtain the first data; First processing: splicing the first data with the historical data of the fire-fighting equipment and recording them as the second data; Define the root node: calculate the information gain ratio of all data features in the second data, and select the data feature with the maximum information gain ratio as the root node; First fault classification training: input the second data into the decision tree layer for inference, calculate the residual of the predicted fault type label and the fault type label corresponding to the historical data, build a decision tree based on the calculated residual and the root node, perform model inference, and output the fault prediction probability; Second fault classification training: The fault prediction probability is concatenated with the second data to obtain the third data, and the third data is used as a sample training set and input into the neural network layer to perform model inference, define a feature loss function, and iteratively optimize the neural network layer by minimizing the feature loss function to obtain a trained fault classification sub-model as a new fault classification sub-model; Second classification: taking the second data as the input of the fault classification sub-model and outputting the fault type label corresponding to the second data; Fault identification: The real-time data of fire-fighting equipment is input into the trained fire-fighting equipment fault identification model, and the fault type label of the real-time data is output.

2. The method for identifying faults of intelligent fire fighting equipment according to claim 1, characterized in that: After executing the step of model building and before executing the step of first classification, the method further includes: Time threshold setting: Different equipment usage time labels are set according to the service life of fire-fighting equipment; Fault type label classification: Different fault types are classified according to the fault classification labels of historical data; Define implicit states: Based on the collected historical data, label the historical data with the corresponding equipment usage time tags, match the equipment usage time tags of the historical data with the corresponding fault types, and use the matched data as the implicit states of the hidden Markov sub-model; First iteration training: input historical data into the hidden Markov sub-model for probability calculation to obtain the forward probability and backward probability of the hidden state, recorded as the first probability, update the observation probability matrix according to the first probability, optimize the hidden Markov sub-model, and obtain the optimized hidden Markov sub-model, recorded as the new hidden Markov sub-model.

3. The method for identifying faults of intelligent fire fighting equipment according to claim 2, characterized in that: After the step of setting the execution time threshold and before the step of classifying the fault type label, the method further includes: First judgment: determine whether the historical equipment operation data in the current historical data are all within the preset operation threshold range: If yes, the types of historical equipment operation data in the current historical data are all marked as normal operation data, and the second determination step is performed; If not, the type of historical equipment operation data that exceeds the preset operation threshold range is marked as abnormal operation data, and the type of historical equipment operation data that does not exceed the preset operation threshold range is marked as normal operation data, and the second determination step is performed; Second judgment: judge whether the historical mechanical characteristic data in the current historical data is within the preset characteristic threshold range: If so, the types of historical mechanical characteristic data in the current historical data are all marked as normal characteristic data; If not, the type of historical mechanical characteristic data that exceeds the preset characteristic threshold range is marked as abnormal characteristic data, and the type of historical mechanical characteristic data that does not exceed the preset characteristic threshold range is marked as normal characteristic data; First identification: Match the type of the annotated historical mechanical characteristic data and the type of the historical equipment operation data in the current historical data with the corresponding fault type label, record it as the first matching label, and use the first matching label as the new fault type label of the current historical data.

4. The method for identifying faults of intelligent fire fighting equipment according to claim 2, characterized in that: After executing the model building step and before executing the time threshold setting step, it also includes: Fault digitization: Perform one-hot encoding on the fault classification label to obtain the first label; First screening: calculating the correlation between the historical equipment operation data and the first tag, recording it as the first correlation data, screening out the historical equipment operation data whose first correlation data is higher than the preset correlation state threshold value as the new historical equipment operation data; Second screening: Calculate the correlation between the historical mechanical feature data and the first label, record it as the second correlation data, and screen out the historical mechanical feature data whose second correlation data is higher than the preset correlation feature threshold as the new historical mechanical feature data.

5. The method for identifying faults of intelligent fire fighting equipment according to claim 2, characterized in that: After performing the first classification step and before performing the first processing step, the method further includes: Probability difference calculation: Calculate the probability difference of the same data under different implicit states in the first data, record it as the first probability difference, and use the first probability difference and the first data as the new first data.

6. The method for identifying faults of intelligent fire fighting equipment according to claim 1, characterized in that: The feature loss function is: Among them, n is the number of samples in the sample training set, w j represents the weight of the jth data feature in the third data, x ij represents the jth data feature in the i-th sample, represents the predicted value of the jth data feature in the i-th sample, γ represents the balance coefficient, and y i is the true label of the fault type label of the i-th sample, The predicted fault type label of the i-th sample output by the neural network layer.

7. The method for identifying faults of intelligent fire fighting equipment according to claim 1, characterized in that: After executing the step of the first processing and before executing the step of defining the root node, the method further includes: Third screening: calculating the correlation between the data features in the second data, recorded as third correlation data, and screening out the data feature group whose third correlation data is greater than a preset screening threshold, recorded as the first data group; Feature selection: Use principal component analysis to convert the two data features of each data group in the first data group into fused data features, recorded as fused data features, collect the fused data features of each data group to obtain a fused data feature group, and use the fused data feature group to replace the data of the first data group in the second data to obtain the replaced data as the new second data.

8. A smart fire equipment fault identification system, characterized in that: The system is applicable to the method according to any one of claims 1 to 7, and the system comprises: Data collection module, including real-time data collection module and historical data collection module; Real-time data collection module, used to collect real-time data of fire-fighting equipment; A historical data collection module is used to collect historical data of fire-fighting equipment and fault type labels corresponding to the historical data, wherein the historical data includes historical equipment operation data and historical mechanical characteristic data; A model building module, used to build a fire equipment fault identification model, wherein the fire equipment fault identification model includes a hidden Markov sub-model and a fault classification sub-model; the fault classification sub-model includes a decision tree layer and a neural network layer; The model training module is connected to the data acquisition module and the model building module in communication, and includes a first classification unit, a first processing unit, a root node definition unit, a first fault classification training unit, a second fault classification training unit, and a second classification unit; A first classification unit is used to use the historical data of the fire-fighting equipment and the fault type label corresponding to the historical data as the input of the hidden Markov sub-model to obtain first data; A first processing unit is used to combine the first data with the historical data of the fire-fighting equipment to record the data as second data; A root node unit is defined to calculate the information gain ratio of all data features in the second data, and the data feature with the maximum information gain ratio is selected as the root node; A first fault classification training unit is used to input the second data into a decision tree layer for reasoning, calculate the residual of the predicted fault type label and the fault type label corresponding to the historical data, build a decision tree according to the calculated residual and the root node, perform model reasoning, and output the fault prediction probability; The second fault classification training unit is used to concatenate the fault prediction probability with the second data to obtain third data, use the third data as a sample training set, input the third data into the neural network layer, perform model inference, define a feature loss function, iteratively optimize the neural network layer by minimizing the feature loss function, and obtain a trained fault classification sub-model as a new fault classification sub-model; A second classification unit, used to use the second data as an input of a fault classification sub-model and output a fault type label corresponding to the historical data; The fault identification unit is connected to the data acquisition module and the model building module for inputting the real-time data of the fire-fighting equipment into the trained fire-fighting equipment fault identification model and outputting the fault type label of the real-time data.

Images