Traction cooling unit self-response abnormity early warning method based on heat exchange modeling

Through the method based on heat exchange modeling, the oil temperature and fresh air temperature data of the traction cooling unit are used to construct health indicators and dynamically adjust the threshold, solving the problems of low warning accuracy and poor generalization ability in the existing technology, achieving a higher Early warning accuracy and stability, suitable for different environments and lines.

CN118966043APending Publication Date: 2024-11-15CRRC CHANGCHUN RAILWAY VEHICLES CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202410891244.7
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-07-04
Publication Date
2024-11-15

AI Technical Summary

Technical Problem

In the prior art, when early warning of factors such as filter clogging, faults and changes in cooling system oil temperature of the traction cooling unit, the accuracy of early warning is not high, and the generalization ability depends on a fixed threshold, which makes it impossible to adapt to changes in different environments and lines.

Method used

Based on the method of heat exchange modeling, health indicators are constructed using the inlet and outlet oil temperature and fresh air temperature data of the traction cooling unit, and early warning is performed through self-responsive and self-learning thresholds, and the threshold is dynamically adjusted to improve early warning accuracy and stability.

Benefits of technology

It achieves higher early warning accuracy and stability, overcomes the problems of low early warning accuracy and limited generalization of the existing methods, and can be applied to different environments and lines, reducing resource waste and operation and maintenance costs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118966043A_ABST
    Figure CN118966043A_ABST
Patent Text Reader

Abstract

A traction cooling unit self-response abnormity early warning method based on heat exchange modeling can effectively monitor performance degradation of a traction cooling system and timely and accurately give an early warning. According to the method, state monitoring quantities such as oil temperature change and fresh air temperature of a cooling system are used as input, health indexes are established based on physical modeling in the convective heat exchange process, and early warning is made for cooling performance abnormity of the traction cooling unit through a self-adaptive threshold value. According to the method, the problems of extremely low early warning accuracy caused by insufficient utilization of monitoring information and generalization limitation caused by a fixed threshold value in the existing method are solved, and accurate early warning of the performance abnormity of the traction cooling system is realized. The method has the advantages of being higher in early warning precision, better in stability and higher in generalization.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention belongs to the field of abnormal warning of mechanical equipment, and in particular relates to a self-response abnormal warning method of a traction cooling unit based on heat exchange modeling. Background Art

[0002] Traction transformers or traction converters generate a lot of heat during the running of trains. To ensure their normal operation, this heat will be taken away by their respective cooling units through air cooling. The cooling fan of the traction cooling unit draws cooling air from the outside to act on the radiator to achieve the purpose of cooling. When the filter of the cooling unit is blocked to a certain extent, or the sludge adhesion of the radiator reaches a certain extent, and the cooling unit cannot meet the actual requirements, it will cause the traction system to overheat and cause safety problems, which will have a great impact on the normal operation of the train. At present, various operating units usually use planned maintenance to ensure the normal operation of the traction cooling unit. However, with the different seasons, running lines, and environments, the degree of filter blockage and the degree of sludge adhesion of the radiator are also different. The implementation of planned maintenance without monitoring will cause some train sets to be over-maintained, resulting in a waste of resources, while some train sets will be at risk of under-repair, affecting driving safety.

[0003] Thanks to advanced sensing and communication technologies, train status monitoring data, environmental condition data, and maintenance application data can all be collected. Through in-depth analysis and mining of these data, an abnormal prediction or performance degradation model can be constructed, which is of vital importance to ensuring the safety of train operation and achieving the transition from planned maintenance to condition-based maintenance. However, existing methods fail to fully consider the relationship between factors such as filter blockage, traction cooling unit failure, cooling system oil temperature changes, and fresh air temperature, resulting in low warning accuracy. In addition, existing methods usually rely on predefined thresholds to trigger filter blockage warnings, which requires that thresholds be set empirically according to the specific conditions of different lines or even different suppliers, and remain fixed during use. Therefore, the application of this method is limited and the generalization ability is poor. Summary of the invention

[0004] In order to overcome the shortcomings of the above-mentioned prior art, the purpose of the present invention is to propose a self-responsive abnormal warning method for a traction cooling unit based on heat exchange modeling. The method uses status monitoring data such as the inlet and outlet oil temperature and the fresh air temperature of the cooling unit as input, constructs health indicators based on the physical modeling of the convective heat transfer process, and warns of performance abnormalities of the traction cooling unit through self-responsive and self-learning thresholds; the method has the advantages of higher warning accuracy, better stability, and stronger generalization.

[0005] In order to achieve the above object, the technical solution adopted by the present invention is:

[0006] A traction cooling unit self-response abnormal warning method based on heat exchange modeling includes the following steps:

[0007] 1) The time span S=n·Span used for learning the self-response threshold is set, where n is a positive integer and Span is the time span required for a single monitoring.

[0008] 2) Create table A to store health indicator values.

[0009] 3) Create table B for storing self-response thresholds, and add an initial threshold th thereto, which is selected based on experience.

[0010] 4) Create table C to record the threshold calculation start time, and add the initial test time, that is, the first start time of the algorithm, to it.

[0011] 5) Obtain the monitoring data of the traction cooling system within a monitoring time Span T i ,T o ,T e ,TS,CS, where T i Indicates the inlet oil temperature data of the traction cooling system during this period, T o Indicates the outlet oil temperature of the traction cooling system during this period, T e It indicates the outdoor fresh air temperature data during this time period, TS indicates the train speed during this time period, and CS indicates the cooling fan opening status during this time period.

[0012] 6) Preprocess the raw data: first align the data according to the timestamp, then remove the duplicate timestamps, and finally resample the data. The specific steps are as follows:

[0013] 6.1) Align and splice all data according to timestamps and keep all timestamps to get Where Q is the number of monitoring data obtained after alignment and splicing within the monitoring time.

[0014] 6.2) Removal Duplicate timestamps in the loop are first detected Checks whether the timestamp is repeated and returns a Boolean sequence indicating whether it is repeated. The sequence and The length is the same, where the value at the repeated timestamp position is True, the value at the non-repeated timestamp position is False, and the first timestamp will be marked as False; then according to from Select the row corresponding to the unique timestamp, that is, b i = False, thus obtaining data without duplicate timestamps.

[0015] 7) Perform preliminary data screening based on monitoring data to reduce the impact of operating conditions. The specific steps are as follows:

[0016] 7.1) Get the train speed TS t Data at all times greater than a certain value, which is determined according to the maximum speed of the train, usually 0.7 times the maximum speed. The remaining data that are not taken retain the timestamp and replace the characteristic value with a null value.

[0017] 7.2) Based on the result after 3.1) take out the cooling fan status CS t For the data at all the time when it is turned on, the timestamps of the remaining data that have not been retrieved are retained and the feature values ​​are replaced by null values.

[0018] 8) To resample the data, first Divide the data into fixed time intervals according to the time interval Interval, and then calculate the average value of each variable in each interval; finally, these average values ​​are used as new data points, where, is the number of sample points of the new data.

[0019] 9) Based on Calculate the health index of the traction cooling unit at each moment The specific steps are as follows:

[0020] 9.1) Determine the calculation method of health indicators. The specific steps are as follows:

[0021] (a) Calculate the convective heat transfer of the traction cooling system per unit time at each moment. The expression is as follows:

[0022]

[0023] Where ΔQ t It represents the heat exchanged between the cooling oil and cooling pipe and the air per unit time, h is the convection heat transfer coefficient of the air, S is the surface area of ​​the air in contact with the cooling pipe, and in the present invention refers to the windward surface area of ​​the cooling pipe.

[0024] (b) Determine the convective heat transfer coefficient h of air. The mathematical expression is as follows:

[0025] h=0.023·(Re 0.8 ·Pr 0.4 )

[0026] Wherein, h represents the forced convection heat transfer coefficient of air, Re represents the Reynolds number, and Pr represents the Prandtl number.

[0027] (c) Determine the Reynolds number Re of the air, which describes the flow characteristics of the air and is mathematically expressed as follows:

[0028] Re=ρvL / μ

[0029] Among them, ρ represents the air density, v represents the flow velocity of the air, L represents the characteristic length, that is, the pipe diameter, and μ represents the dynamic viscosity of the air.

[0030] (d) Determine the Prandtl number of air, which reflects the influence of fluid physical properties on the convective heat transfer process. The mathematical expression is as follows:

[0031] Pr=C p μ / k

[0032] Among them C p represents the constant pressure specific heat capacity of air, μ represents the dynamic viscosity of air, and k represents the thermal conductivity

[0033] (e) Determine the relationship between heat and cooling oil temperature change. The mathematical expression is as follows:

[0034] ΔQ t =m oil ·C p,oil ΔT oil

[0035] Among them, ΔQ t Indicates the heat exchanged between the cooling oil and cooling pipe and the air per unit time, m oil Indicates the mass of oil flowing per unit time, C p,oil Indicates the specific heat capacity of the oil. If the dirt adheres to the heat dissipation copper tube, it can be equivalent to C p,oil Increase, that is, more energy is required to escape to reduce the temperature by the same amount, ΔT oil Indicates the change in oil temperature per unit time.

[0036] (f) Combining (a) to (e), the relationship between air velocity, oil specific heat capacity, inlet and outlet oil temperature and outdoor fresh air temperature is obtained, which can be expressed mathematically as follows:

[0037]

[0038] in, is a constant, which can be approximated as a constant during the operation of the cooling system, ΔT oil Indicates the change in oil temperature per unit time, v: air velocity, in meters per second (m / s); the smaller the air velocity, the greater the degree of filter blockage, C p,oil Indicates the specific heat capacity of the oil. If the dirt adheres to the heat dissipation copper tube, it can be equivalent to C p,oil Increase, that is, more energy is required to escape to reduce the temperature by the same amount, that is The smaller the value, the worse the cooling system. p,oilcan jointly characterize the working condition of the cooling system, that is The larger the cooling system is, the worse the cooling performance is. Therefore, the method of the present invention will Serves as an indicator of the health of the traction cooling unit.

[0039] (g) Integrate and transform the mathematical expression in (f) to obtain the health index The calculation expression is as follows:

[0040]

[0041] in It can be approximated as a constant and replaced by 1 in actual use, that is,

[0042] 9.2) After determining the calculation method of health indicators, based on Determine the health indicators of the traction cooling unit at each moment The calculation method and specific steps are as follows:

[0043] (a) Calculate the health index of the traction cooling unit at each moment. The calculation expression is as follows:

[0044]

[0045] (b) Use the scaling function to further scale the health indicators to expand the separability of normal and abnormal states. The calculation expression is as follows:

[0046]

[0047] in is the scaled health index value, ζ is the hyperparameter that controls the scale scaling, which is calculated by calculating all HI when the cooling system of multiple trains is working normally. t The average value is obtained.

[0048] 10) Calculate the average value of the health indicator within the monitoring time span The calculation expression is as follows:

[0049]

[0050] 11) Compare with the latest threshold in Table B. If it exceeds the threshold, a warning message is given and pushed to the operation and maintenance system.

[0051] 12) As a health indicator within the monitoring time span, its value is added to Table A.

[0052] 13) Determine the latest health index HI compared to the previous non-empty health index Whether the reduction is greater than 70%, that is, Is it true? If so, proceed to 14), otherwise write the latest threshold calculation time into Table C.

[0053] 14) Determine whether the length of table C is greater than 0. If the length of table C is greater than 0, execute 15). Otherwise, take the previous threshold in table B as the latest threshold and add it to table B.

[0054] 15) Take the time from the latest threshold calculation in Table C to the current time as the interval, calculate whether the number of non-empty threshold points in Table B within the time interval is less than or equal to n. If so, execute 16), otherwise take the previous threshold in Table B as the latest threshold and add it to Table B.

[0055] 16) Calculate the mean μ and variance σ of all health indicators HI recorded in Table A, and use μ+mσ as the latest threshold and add it to Table B.

[0056] 17) Repeat 5) to 16) until the algorithm is deactivated.

[0057] The beneficial effects of the present invention are:

[0058] The method of the present invention can use the state monitoring quantities such as the inlet and outlet oil temperatures and the fresh air temperature of the traction cooling unit, construct a health indicator based on the physical modeling of the convective heat transfer process, and evaluate the performance of the traction cooling unit and give an abnormal warning through a self-responsive and self-learning threshold. The method of the present invention overcomes the problem of extremely low warning accuracy caused by insufficient utilization of monitoring information in the existing methods, and at the same time overcomes the problem of limited generalization of the existing methods due to the use of fixed thresholds, thereby achieving accurate early warning of abnormal performance of the traction cooling unit. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 Flow chart of the method of the present invention.

[0060] Figure 2 It is the main status monitoring quantity of vehicle group A on the first day of monitoring.

[0061] Figure 3 These are the early warning results of the method of the present invention for three vehicle groups.

[0062] Figure 4 The following is the early warning results of three vehicle groups using the comparison method. DETAILED DESCRIPTION

[0063] The present invention will be further described in detail below in conjunction with the accompanying drawings.

[0064] Reference Figure 1, a traction cooling unit self-response abnormal warning method based on heat exchange modeling, comprising the following steps:

[0065] 1) The time span S=n·Span used for learning the self-response threshold is set, where n is a positive integer and Span is the time span required for a single monitoring.

[0066] 2) Create table A to store health indicator values.

[0067] 3) Create table B for storing self-response thresholds, and add an initial threshold th thereto, which is selected based on experience.

[0068] 4) Create table C to record the threshold calculation start time, and add the initial test time, that is, the first start time of the algorithm, to it.

[0069] 5) Obtain the monitoring data of the traction cooling system within a monitoring time Span T i ,T o ,T e ,TS,CS, where T i Indicates the inlet oil temperature data of the traction cooling system during this period, T o Indicates the outlet oil temperature of the traction cooling system during this period, T e It indicates the outdoor fresh air temperature data during this time period, TS indicates the train speed during this time period, and CS indicates the cooling fan opening status during this time period.

[0070] 6) Preprocess the raw data: first align the data according to the timestamp, then remove the duplicate timestamps, and finally resample the data. The specific steps are as follows:

[0071] 6.1) Align and splice all data according to timestamps and keep all timestamps to get Where Q is the number of monitoring data obtained after alignment and splicing within the monitoring time.

[0072] 6.2) Removal Duplicate timestamps in the loop are first detected Checks whether the timestamp is repeated and returns a Boolean sequence indicating whether it is repeated. The sequence and The length is the same, where the value at the repeated timestamp position is True, the value at the non-repeated timestamp position is False, and the first timestamp will be marked as False; then according to from Select the row corresponding to the unique timestamp, that is, b i = False, thus obtaining data without duplicate timestamps.

[0073] 7) Perform preliminary data screening based on monitoring data to reduce the impact of operating conditions. The specific steps are as follows:

[0074] 7.1) Get the train speed TS t Data at all times greater than a certain value, which is determined according to the maximum speed of the train, usually 0.7 times the maximum speed. The remaining data that are not taken retain the timestamp and replace the characteristic value with a null value.

[0075] 7.2) Based on the result after 3.1) take out the cooling fan status CS t For the data at all the time when it is turned on, the timestamps of the remaining data that have not been retrieved are retained and the feature values ​​are replaced by null values.

[0076] 8) To resample the data, first Divide the data into fixed time intervals according to the time interval Interval, and then calculate the average value of each variable in each interval; finally, these average values ​​are used as new data points, where, is the number of sample points of the new data.

[0077] 9) Based on Calculate the health index of the traction cooling unit at each moment The specific steps are as follows:

[0078] 9.1) Determine the calculation method of health indicators. The specific steps are as follows:

[0079] (a) Calculate the convective heat transfer of the traction cooling system per unit time at each moment. The expression is as follows:

[0080]

[0081] Where ΔQ t It represents the heat exchanged between the cooling oil and cooling pipe and the air per unit time, h is the convection heat transfer coefficient of the air, S is the surface area of ​​the air in contact with the cooling pipe, and in the present invention refers to the windward surface area of ​​the cooling pipe.

[0082] (b) Determine the convective heat transfer coefficient h of air. The mathematical expression is as follows:

[0083] h=0.023·(Re 0.8 ·Pr 0.4 )

[0084] Wherein, h represents the forced convection heat transfer coefficient of air, Re represents the Reynolds number, and Pr represents the Prandtl number.

[0085] (c) Determine the Reynolds number Re of the air, which describes the flow characteristics of the air and is mathematically expressed as follows:

[0086] Re=ρvL / μ

[0087] Among them, ρ represents the air density, v represents the flow velocity of the air, L represents the characteristic length, that is, the pipe diameter, and μ represents the dynamic viscosity of the air.

[0088] (d) Determine the Prandtl number of air, which reflects the influence of fluid physical properties on the convective heat transfer process. The mathematical expression is as follows:

[0089] Pr=C p μ / k

[0090] Among them C p represents the constant pressure specific heat capacity of air, μ represents the dynamic viscosity of air, and k represents the thermal conductivity

[0091] (e) Determine the relationship between heat and cooling oil temperature change. The mathematical expression is as follows:

[0092] ΔQ t =m oil ·C p,oil ΔT oil

[0093] Among them, ΔQ t Indicates the heat exchanged between the cooling oil and cooling pipe and the air per unit time, m oil Indicates the mass of oil flowing per unit time, C p,oil Indicates the specific heat capacity of the oil. If the dirt adheres to the heat dissipation copper tube, it can be equivalent to C p,oil Increase, that is, more energy is required to escape to reduce the temperature by the same amount, ΔT oil Indicates the change in oil temperature per unit time.

[0094] (f) Combining (a) to (e), the relationship between air velocity, oil specific heat capacity, inlet and outlet oil temperature and outdoor fresh air temperature is obtained, which can be expressed mathematically as follows:

[0095]

[0096] in, is a constant, which can be approximated as a constant during the operation of the cooling system, ΔT oil Indicates the change in oil temperature per unit time, v: air velocity, in meters per second (m / s); the smaller the air velocity, the greater the degree of filter blockage, C p,oil Indicates the specific heat capacity of the oil. If the dirt adheres to the heat dissipation copper tube, it can be equivalent to C p,oil Increase, that is, more energy is required to escape to reduce the temperature by the same amount, that is The smaller the value, the worse the cooling system. p,oilcan jointly characterize the working condition of the cooling system, that is The larger the cooling system is, the worse the cooling performance is. Therefore, the method of the present invention will Serves as an indicator of the health of the traction cooling unit.

[0097] (g) Integrate and transform the mathematical expression in (f) to obtain the health index The calculation expression is as follows:

[0098]

[0099] in It can be approximated as a constant and replaced by 1 in actual use, that is,

[0100] 9.2) After determining the calculation method of health indicators, based on Determine the health indicators of the traction cooling unit at each moment The calculation method and specific steps are as follows:

[0101] (a) Calculate the health index of the traction cooling unit at each moment. The calculation expression is as follows:

[0102]

[0103] (b) Use the scaling function to further scale the health indicators to expand the separability of normal and abnormal states. The calculation expression is as follows:

[0104]

[0105] in is the scaled health index value, ζ is the hyperparameter that controls the scale scaling, which is calculated by calculating all HI when the cooling system of multiple trains is working normally. t The average value is obtained.

[0106] 10) Calculate the average value of the health indicator within the monitoring time span The calculation expression is as follows:

[0107]

[0108] 11) Compare with the latest threshold in Table B. If it exceeds the threshold, a warning message is given and pushed to the operation and maintenance system.

[0109] 12) As a health indicator within the monitoring time span and add its value to table A

[0110] 13) Determine the latest health indicators Compared to the previous non-empty health indicator Whether the reduction is greater than 70%, that is, Is it true? If so, proceed to 14), otherwise write the latest threshold calculation time into Table C.

[0111] 14) Determine whether the length of table C is greater than 0. If the length of table C is greater than 0, execute 15). Otherwise, take the previous threshold in table B as the latest threshold and add it to table B.

[0112] 15) Take the time from the latest threshold calculation in Table C to the current time as the interval, calculate whether the number of non-empty threshold points in Table B within the time interval is less than or equal to n. If so, execute 16), otherwise take the previous threshold in Table B as the latest threshold and add it to Table B.

[0113] 16) Calculate all health indicators recorded in Table A The mean μ and variance σ of , and μ+mσ are taken as the latest threshold and added to Table B.

[0114] 17) Repeat 5) to 16) until the algorithm is deactivated.

[0115] Embodiment: The effectiveness of the method of the present invention is verified based on the monitoring data of the cooling unit of the traction transformer of a certain type of EMU in my country. The monitoring data involved in this embodiment comes from three different EMUs. During the two-month monitoring period, one group of EMUs operated normally, while the other two groups experienced cooling oil overtemperature failures. Subsequent return to the depot for inspection confirmed that these failures were caused by blockage of the cooling system filter. The status data collected during the monitoring period include train speed, outdoor temperature, inlet and outlet oil temperature of the traction transformer cooling unit, high and low speed status of the cooling fan, etc. The sampling frequency of all these data is 30 seconds. Table 1 shows an overview of the data, and Figure 2 The main status monitoring values ​​of train A on the first day of monitoring are shown in Table 2. The hyperparameter settings of the method of the present invention are shown in Table 2. The health monitoring and performance abnormality warning of the three EMUs are performed by applying the method of the present invention. The warning results are shown in Figure 3 As shown. Figure 3It can be clearly seen that the method of the present invention successfully models the performance degradation process of the traction cooling unit and can accurately evaluate the performance. In addition, the adaptive threshold of the present method can be learned and updated accurately and flexibly without manual adjustment. Therefore, the method of the present invention issued a timely warning before the traction cooling unit of the two EMUs failed, and no false alarms were generated during the period, which proves the ability of the present method to effectively monitor the performance degradation of the traction cooling system and issue timely and accurate warnings. In order to verify the superiority of the method of the present invention, a comparative method was further constructed to perform health monitoring and abnormal performance warnings on the three train sets. Unlike the inventive method, the comparative method adopts the weighted index method commonly used in the prior art to construct health indicators, and uses fixed thresholds during monitoring. The warning results of the comparative method are as follows. Figure 4 As shown. Figure 4 It can be seen that the comparison method produces serious premature false alarms during the monitoring process, which may mislead on-site operation and maintenance personnel and increase operation and maintenance costs. In addition, the threshold of the comparison method depends on historical experience and manual settings, which is not conducive to the maintenance of the algorithm after deployment, nor is it conducive to the generalization application of the algorithm. In comparison, the method of the present invention shows significant advantages. It can not only model and evaluate the performance degradation of the traction cooling system more accurately, but also adaptively learn and update the warning threshold, avoiding the need for manual intervention. This automated threshold adjustment mechanism improves the flexibility and reliability of the early warning system and reduces the possibility of false alarms, thereby reducing operation and maintenance costs in practical applications and improving the operation and maintenance efficiency of the overall system. In addition, the powerful generalization ability of the method of the present invention enables it to be applicable to different EMUs and usage environments without the need to readjust parameters for each specific scenario, which greatly simplifies maintenance work in actual operation and enhances the applicability and promotion value of the method.

[0116] Table 1

[0117]

[0118] Table 2

[0119]

[0120] Through the above abnormal warning results of the traction cooling system and the performance comparison with the comparative methods, it can be found that the method of the present invention can use the state monitoring quantities such as the oil temperature change and the fresh air temperature of the traction cooling system, establish the health index based on the physical modeling of the convective heat transfer process, and use the adaptive threshold to warn of the abnormal cooling performance of the traction cooling unit; it overcomes the problems of the existing methods such as insufficient utilization of monitoring information resulting in extremely low warning accuracy and limited generalization based on fixed thresholds, and realizes accurate warning of abnormal performance of the traction cooling system.

Claims

1. A self-response abnormal warning method for a traction cooling unit based on heat exchange modeling, characterized in that: The following steps are involved: 1) The time span S=n·Span used for learning the self-response threshold is set, where n is a positive integer and Span is the time span required for a single monitoring. 2) Create table A to store health indicator values. 3) Create table B for storing self-response thresholds, and add an initial threshold th thereto, which is selected based on experience. 4) Create table C to record the threshold calculation start time, and add the initial test time, that is, the first start time of the algorithm, to it. 5) Obtain the monitoring data of the traction cooling system within a monitoring time Span T i ,T o ,T e ,TS,CS, where T i Indicates the inlet oil temperature data of the traction cooling system during this period, T o Indicates the outlet oil temperature of the traction cooling system during this period, T e It indicates the outdoor fresh air temperature data during this time period, TS indicates the train speed during this time period, and CS indicates the cooling fan opening status during this time period. 6) Preprocess the raw data: first align the data according to the timestamp, then remove the duplicate timestamps, and finally resample the data. The specific steps are as follows: 6.1) Align and splice all data according to timestamps and keep all timestamps to get Where Q is the number of monitoring data obtained after alignment and splicing within the monitoring time. 6.2) Removal Duplicate timestamps in the loop are first detected Checks whether the timestamp is repeated and returns a Boolean sequence indicating whether it is repeated. The sequence and The length is the same, where the value at the repeated timestamp position is True, the value at the non-repeated timestamp position is False, and the first timestamp will be marked as False; then according to from Select the row corresponding to the unique timestamp, that is, b i = False, thus obtaining data without duplicate timestamps. 7) Perform preliminary data screening based on monitoring data to reduce the impact of operating conditions. The specific steps are as follows: 7.1) Get the train speed TS t Data at all times greater than a certain value, which is determined according to the maximum speed of the train, usually 0.7 times the maximum speed. The remaining data not taken retains the timestamp and replaces the characteristic value with a null value. 7.2) Based on the result after 3.1) take out the cooling fan status CS t For the data at all the time when it is turned on, the timestamps of the remaining data that have not been retrieved are retained and the feature values ​​are replaced by null values. 8) To resample the data, first Divide the data into fixed time intervals according to the time interval Interval, and then calculate the average value of each variable in each interval; finally, these average values ​​are used as new data points, where, is the number of sample points of the new data. 9) Based on Calculate the health index of the traction cooling unit at each moment The specific steps are as follows: 9.1) Determine the calculation method of health indicators. The specific steps are as follows: (a) Calculate the convective heat transfer of the traction cooling system per unit time at each moment. The expression is as follows: Where ΔQ t It represents the heat exchanged between the cooling oil and cooling pipe and the air per unit time, h is the convection heat transfer coefficient of the air, S is the surface area of ​​the air in contact with the cooling pipe, and in the present invention refers to the windward surface area of ​​the cooling pipe. (b) Determine the convective heat transfer coefficient h of air. The mathematical expression is as follows: h=0.023·(Re 0.8 ·Pr 0.4 , Wherein, h represents the forced convection heat transfer coefficient of air, Re represents the Reynolds number, and Pr represents the Prandtl number. (c) Determine the Reynolds number Re of the air, which describes the flow characteristics of the air and is mathematically expressed as follows: Re=ρvL / μ Among them, ρ represents the air density, v represents the flow velocity of the air, L represents the characteristic length, that is, the pipe diameter, and μ represents the dynamic viscosity of the air. (d) Determine the Prandtl number of air, which reflects the influence of fluid physical properties on the convective heat transfer process. The mathematical expression is as follows: Pr=C p μ / k Among them C p represents the constant pressure specific heat capacity of air, μ represents the dynamic viscosity of air, and k represents the thermal conductivity (e) Determine the relationship between heat and cooling oil temperature change. The mathematical expression is as follows: ΔQ t =m oil ·C p,oil ·ΔT oil Among them, ΔQ t Indicates the heat exchanged between the cooling oil and cooling pipe and the air per unit time, m oil Indicates the mass of oil flowing per unit time, C p,oil Indicates the specific heat capacity of the oil. If the dirt adheres to the heat dissipation copper tube, it can be equivalent to C p,oil Increase, that is, more energy is required to escape to reduce the temperature by the same amount, ΔT oil Indicates the change in oil temperature per unit time. (f) Combining (a) to (e), the relationship between air velocity, oil specific heat capacity, inlet and outlet oil temperature and outdoor fresh air temperature is obtained, which can be expressed mathematically as follows: in, is a constant, which can be approximated as a constant during the operation of the cooling system, ΔT oil Indicates the change in oil temperature per unit time, v: air velocity, in meters per second (m / s); the smaller the air velocity, the greater the degree of filter blockage, C p,oil Indicates the specific heat capacity of the oil. If the dirt adheres to the heat dissipation copper tube, it can be equivalent to C p,oil Increase, that is, more energy is required to escape to reduce the temperature by the same amount, that is The smaller the value, the worse the cooling system. p,oil can jointly characterize the working condition of the cooling system, that is The larger the cooling system is, the worse the cooling performance is. Therefore, the method of the present invention will Serves as an indicator of the health of the traction cooling unit. (g) Integrate and transform the mathematical expression in (f) to obtain the health index The calculation expression is as follows: in It can be approximated as a constant and replaced by 1 in actual use, that is, 9.2) After determining the calculation method of health indicators, based on Determine the health indicators of the traction cooling unit at each moment The calculation method and specific steps are as follows: (a) Calculate the health index of the traction cooling unit at each moment. The calculation expression is as follows: (b) Use the scaling function to further scale the health indicators to expand the separability of normal and abnormal states. The calculation expression is as follows: in is the scaled health index value, ζ is the hyperparameter that controls the scale scaling, which is calculated by calculating all HI when the cooling system of multiple trains is working normally. t The average value is obtained. 10) Calculate the average value of the health indicator within the monitoring time span The calculation expression is as follows: 11) Compare with the latest threshold in Table B. If it exceeds the threshold, a warning message is given and pushed to the operation and maintenance system. 12) As a health indicator within the monitoring time span and add its value to table A 13) Determine the latest health indicators Compared to the previous non-empty health indicator Whether the reduction is greater than 70%, that is, Is it true? If so, proceed to 14), otherwise write the latest threshold calculation time into Table C. 14) Determine whether the length of table C is greater than 0. If the length of table C is greater than 0, execute 15). Otherwise, take the previous threshold in table B as the latest threshold and add it to table B. 15) Take the time from the latest threshold calculation in Table C to the current time as the interval, calculate whether the number of non-empty threshold points in Table B within the time interval is less than or equal to n. If so, execute 16), otherwise take the previous threshold in Table B as the latest threshold and add it to Table B. 16) Calculate all health indicators recorded in Table A The mean μ and variance σ of , and μ+mσ are taken as the latest threshold and added to Table B. 17) Repeat 5) to 16) until the algorithm is deactivated.