Diesel engine destructive damage monitoring and early warning method based on thermal data driving

Abstract

The application provides a diesel engine destructive damage monitoring and early warning method based on thermal data driving, which comprises the following steps: diesel engine bench destructive damage sign opportunity capturing and sensitive parameter identification; and diesel engine bench destructive damage early warning. The application has the beneficial effects that: the human intervention process of sensitive index and detection opportunity control marking is avoided, the warning effect has higher repeatability and extremely high reliability; the early warning method provided by the application reduces the working strength of the operator, and can continuously expand the subsequent destructive damage identification knowledge, meet the safety warning of various component performance verification, and has extremely high value for the lightweight development of the diesel engine.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of diesel engine bench monitoring, and particularly relates to a diesel engine destructive damage monitoring and early warning method based on thermal data driving. BACKGROUND

[0002] Bench test is an indispensable link for diesel engine research and performance identification, and a new diesel engine needs to be identified through multiple rounds of bench performance and durability verification to determine the technical state of the design goal of low cost, high power and high reliability.

[0003] Problems existing in the prior art:

[0004] (1) In order to prevent the diesel engine from being damaged, the diesel engine controller has certain water temperature, oil temperature, exhaust temperature and oil pressure protection measures, but in order to meet the high reliability and low cost target, the number of sensors installed on board is extremely limited, the diesel engine is not sensitive to the abnormal running state, especially in the high risk limit verification process of high transient state and overload, it is difficult to effectively warn, and the serious fault of the research test occurs frequently; (2) The current diesel engine thermal collection bench has complete thermal detection parameter collection and alarm functions, but its core main function is mainly for process control of the running condition, including adjusting the water temperature, oil temperature, oil supply and load in the test according to the test specification, the safety alarm limit depends on the manual experience setting, and in order to meet the normal operation of the diesel engine in the whole working condition range, the fault alarm tolerance limit is relatively wide, which cannot reflect the time-varying deterioration state of the diesel engine after entering the working condition, the response warning degree of the high transient fault is not high, and the fault alarm opportunity is late, which cannot protect the diesel engine in time; (3) In the verification of the new diesel engine of the conventional main engine factory, experienced test personnel are usually configured to test the safety of the equipment and the diesel engine, but the abnormal process is random, and it is difficult for the personnel to control the timely shutdown opportunity, the fault is exposed in the late stage, which has caused serious loss, and the high mechanical load factor of the light weight design of the diesel engine makes it difficult for the personnel to respond to the safety monitoring window period. SUMMARY

[0005] Therefore, the present application aims to provide a diesel engine destructive damage monitoring and early warning method based on thermal data driving to solve at least one of the above problems existing in the prior art.

[0006] To achieve the above purpose, the technical scheme of the present application is as follows:

[0007] The diesel engine destructive damage monitoring and early warning method based on thermal data driving comprises the following steps:

[0008] S1, capturing the sign opportunity of diesel engine bench destructive damage and identifying sensitive parameters;

[0009] S2, diesel engine bench destructive damage early warning;

[0010] In step S1, the signs of diesel engine bench destructive damage are captured and sensitive parameters are identified, including:

[0011] S11, thermal data collection and arrangement to form a data resource;

[0012] S12, based on the data resource, data preprocessing is performed to form a data sample;

[0013] S13, based on the thermal data reflecting time of destructive failure, a moving window function is constructed;

[0014] S14, based on the data sample, the characteristic value is calculated to form a data sample matrix;

[0015] S15, based on the moving window function and the data sample matrix, data set division is performed to form a training set and a test set;

[0016] S16, based on the training set and the test set, random forest model training is performed, and the test set prediction accuracy is obtained;

[0017] S17, based on the test set prediction accuracy, the test set accuracy is determined;

[0018] S18, based on the random forest model, sensitive parameters are extracted;

[0019] In step S2, the diesel engine bench destructive damage early warning includes:

[0020] S21, diesel engine full working condition data preparation;

[0021] S22, based on the diesel engine full working condition data, the test data working condition is classified;

[0022] S23, based on the test data working condition of step S22, random forest model construction and training are performed, and a random forest prediction model is output;

[0023] S24, based on the random forest prediction model of step S23, test bench model deployment is performed;

[0024] S25, during diesel engine operation, a warning set is collected;

[0025] S26, whether the warning set is equal to the destructive damage knowledge set, if yes, jump to step S25, otherwise, proceed to the next step;

[0026] S27, warning and driving bench emergency action position.

[0027] Further, in step S11, the thermal data collection and arrangement form a data resource, including:

[0028] Collecting historical inventory fault process segment data, including test bench synchronous data collection and auxiliary explosion pressure, supercharger speed and smoke data collected by mobile detection instruments;

[0029] Based on the damage site combination when the fault occurs, classification is performed to form a data resource.

[0030] Further, in step S12, based on the data resource, data preprocessing is performed to form a data sample, including:

[0031] Unit transformation is performed on the data resource, abnormal values are removed using the median value method, and missing data items are supplemented with linear interpolation to form a data sample.

[0032] Further, in step S13, based on the thermal data reflecting time of destructive failure, a moving window function is constructed, including:

[0033] The thermal data reflecting time of destructive failure is multiplied by a set multiple as the initial size of the window;

[0034] The median of the window is used as the classification boundary point to construct a classification function.

[0035] Further, in step S14, based on the data sample, the characteristic value is calculated to form a data sample matrix, including:

[0036] The data sample of step S12 is extracted according to the time increment value change speed, and the value change acceleration is combined with the original data to form a data set;

[0037] Each thermal parameter in the data set is segmented according to the constructed window size;

[0038] The segmentation process is intercepted by setting the sliding window to form a data sample matrix.

[0039] Further, in step S15, based on the moving window function and the data sample matrix, the data set is divided to form a training set and a test set, including:

[0040] Based on the moving window function, the target training data is obtained, combined with the data sample matrix, the sample order sequence is shuffled, and the training set and the test set are divided according to the set method.

[0041] Further, in step S16, based on the training set and the test set, the random forest model is trained, and the test set prediction accuracy is obtained, including:

[0042] The thermal parameter data in the training set and the test set is normalized;

[0043] The normalized training set data is substituted into the random forest model, and the moving window function is trained as the training target;

[0044] The test set is tested to obtain the prediction accuracy of the test set.

[0045] Further, in step S17, based on the prediction accuracy of the test set, the test set accuracy is determined, including:

[0046] Based on the prediction accuracy of the test set, the adaptability of the moving window function is determined; if the accuracy is less than the set value, the window size is increased, and the processing is restarted in step S14 until the accuracy is equal to the set value.

[0047] Further, in step S18, based on the random forest model, the sensitive parameter is extracted, including:

[0048] Based on the random forest model, the significance parameter is extracted as the monitoring sensitive parameter of subsequent damage; the timing significance deviation is set as the sign of damage.

[0049] Further, in step S21, the diesel engine full working condition data is prepared, including:

[0050] A set of historical operating universal characteristic data of the diesel engine is sorted out;

[0051] In step S22, the test data working condition classification includes:

[0052] According to the form of the speed-load map, a certain number of segmented working conditions are divided;

[0053] In step S23, the random forest model construction and training includes:

[0054] Random model construction; training data set construction; output prediction model;

[0055] In step S24, the test bench model deployment includes:

[0056] Random forest prediction model deployment;

[0057] Working condition classification map query;

[0058] A certain number of parallel loop random forest prediction models are arranged;

[0059] The running process working condition classification query map is arranged.

[0060] Compared with the prior art, the diesel engine destructive damage monitoring and early warning method based on thermal data driving has the following beneficial effects:

[0061] (1) The application provides a diesel engine destructive damage monitoring and early warning method based on thermal data driving, which is used for rapid monitoring and early warning of diesel engine destructive failure in a test bench, improves the safety of the diesel engine test bench, and solves the problem that in the traditional verification process, the operator with rich verification experience is used as the test master system, due to the personnel fatigue factor in the continuous monitoring process, and the difference in the timeliness of personnel perception and operation, and the mechanical load of the main working parts of the diesel engine is generally high, the failure duration time is short, and the personnel from monitoring experience judgment to effective action response cannot meet the rapid disposal of the test process.

[0062] (2) In order to advance the test bench early warning opportunity, accurately determine the sensitivity of the thermal parameter, avoid the influence of artificial experience on the determination of the parameter, a sliding window function and a random forest model are constructed, the destructive damage detection opportunity mark position is evaluated according to the index, and the sensitive parameter under the corresponding detection opportunity is further determined, so that the problem of failure and scrapping in the verification of the lightweight, high-strength and high-value diesel engine is avoided.

[0063] (3) In order to deploy the early warning strategy in the test bench and have the destructive damage early warning pushing function, the application provides a diesel engine bench destructive damage early warning method, which encodes the thermal parameter detection knowledge and the deviation amount, and reports the early warning in real time through the test bench round monitoring mode, so that the subsequent identification model can be continuously increased, and the test bench process control system is met.

[0064] (4) In order to solve the problem that more state abnormal data training is needed in the traditional process, the application utilizes the deviation characteristics of normal working parameters, combines the known destructive damage knowledge to realize the abnormal state early warning process, the training prediction data source is relatively easy to obtain, the generated abnormal state early warning process also has combination, avoids the misoperation introduced by single parameter abnormality, reduces the false alarm rate of the test bench, and provides a reliable method for reliable development of the test process.

[0065] (5) The application provides a diesel engine destructive damage monitoring and early warning method based on thermal data driving, solves the conversion of the inventory thermal parameter to the test bench monitoring and early warning, reduces the demand for high-risk test personnel working strength, has the early warning coding process, meets the test bench process control deployment and positioning information pushing function, makes the operator quickly remove the abnormal state according to the early warning information, and decides the subsequent test progress, has a positive effect on the diesel engine performance and reliability verification work. At the same time, based on the sliding window mark fault process, the prediction accuracy determination criterion can accurately grasp the abnormal state generation opportunity and sensitive parameter, reduce the artificial experience dependence, and the generated abnormal state identification knowledge has obvious effect. BRIEF DESCRIPTION OF DRAWINGS

[0066] The accompanying drawings, which form a part of this application, are included to provide a further understanding of the application, illustrate the preferred embodiments of the application and assist in

[0067] Figure 1 Flowchart of diesel engine bench destructive damage precursor opportunity capture and sensitive parameter identification process according to an embodiment of the application;

[0068] Figure 2 Flowchart of diesel engine bench destructive damage early warning process according to an embodiment of the application;

[0069] Figure 3 Flowchart of non-destructive damage warning (heat exchanger leakage shutdown) according to an embodiment of the application;

[0070] Figure 4 Flowchart of destructive damage warning (burning shoe shaft shutdown) according to an embodiment of the application;

[0071] Figure 5 Flowchart of acquisition system working exception acquisition data point according to an embodiment of the application;

[0072] Figure 6 Flowchart of IQR criterion rejection and re-linear interpolation according to an embodiment of the application;

[0073] Figure 7 Flowchart of moving window function result according to an embodiment of the application;

[0074] Figure 8 Flowchart of sliding window construction new sequence according to an embodiment of the application;

[0075] Figure 9 Flowchart of merged target sequence according to an embodiment of the application;

[0076] Figure 10 Flowchart of effect of initial default window size according to an embodiment of the application;

[0077] Figure 11 Flowchart of effect of adjusted window size according to an embodiment of the application;

[0078] Figure 12 Flowchart of best recognition opportunity according to an embodiment of the application;

[0079] Figure 13 Flowchart of sensitive parameter identification according to an embodiment of the application;

[0080] Figure 14 Flowchart of diesel engine normal operation universal characteristic oil consumption rate curve according to an embodiment of the application;

[0081] Figure 15 Test data working condition classification schematic diagram for the embodiment of the application;

[0082] Figure 16 Oil consumption rate training parameter schematic diagram for the embodiment of the application;

[0083] Figure 17 Bench normal working oil consumption rate data test parameter schematic diagram for the embodiment of the application. DETAILED DESCRIPTION

[0084] It should be noted that the embodiments in the application and the features in the embodiments can be combined with each other without conflict.

[0085] In the description of the application, it should be understood that the terms "center", "longitudinal", "transverse", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer" and the like indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the application and simplifying the description, and do not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the application. In addition, the terms "first", "second" and the like are only for descriptive purposes and cannot be understood as indicating or implying relative importance or implicitly indicating the number of technical features indicated. Therefore, the features defined as "first", "second" and the like can explicitly or implicitly include one or more of the features. In the description of the application, unless otherwise specified, the meaning of "a plurality of" is two or more.

[0086] In the description of the application, it should be noted that unless otherwise explicitly specified and limited, the terms "mounting", "connection", "connection" should be understood broadly, for example, it can be fixed connection, or detachable connection, or integral connection; it can be mechanical connection, or electrical connection; it can be directly connected, or indirectly connected through an intermediate medium, or the communication inside two elements. For those skilled in the art, the specific meaning of the above terms in the application can be understood through specific circumstances.

[0087] The application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0088] As Figure 1 shown, a diesel engine bench destructive damage symptom opportunity capture and sensitive parameter identification method based on thermal data driving is as follows:

[0089] A1: Thermal data collection and arrangement with state

[0090] Collect historical inventory fault process segment data, including synchronous acquisition data of test bench and auxiliary explosion pressure, supercharger speed, smoke and other data collected by mobile detection instrument, classify based on damage site combination when fault occurs, and form data resources for subsequent identification.

[0091] A2: Data preprocessing

[0092] Unit conversion is performed on the data resources collected in step A1 to avoid parameter variables less than 1, data outliers are removed using the median value method, deviations introduced by unstable acquisition equipment or reading of detection personnel during the test process are solved, linear interpolation is used to supplement data for missing data items, and data integrity is ensured.

[0093] A3: Constructing a moving window function

[0094] Three times the reflection time of the destructive fault of the diesel engine thermal data are taken as the initial size of the window, and a 2-classification function is constructed with the median of the window as the classification boundary. The normal operation process before the fault occurs is marked as 0, and the fault occurs after the thermal data has a perception feature marked as 1.

[0095] A4: Feature value calculation

[0096] The data samples after data preprocessing in step A2 are extracted according to the time increment, the value change speed and the value change acceleration, and are combined with the original data in step A2 to form a data set. Based on the constructed window size, each thermal parameter in the data set formed in step A2 is segmented, and the segmentation process is intercepted according to the sliding window of step A1 to form a data sample matrix of each thermal parameter.

[0097] A5: Data set division

[0098] The results of the moving window function are taken as the target training data, combined with the data sample matrix constructed in step A4, the sample order sequence is shuffled, and the training set and the test set of the random forest model are divided according to the 7:3 mode.

[0099] A6: Random forest model training

[0100] The thermal parameter data in the training set and test set data obtained in step A5 are normalized, the normalized training set data is substituted into the random forest model, and the data of the moving window function is trained as the training target. Further test using the test set to obtain the prediction accuracy of the test set;

[0101] A7: Test set accuracy determination

[0102] According to the accuracy of the test set obtained in step A6, the window function adaptability is determined. If the accuracy is less than 100%, the window size of the window function in step A3 is increased, and the processing is restarted in step A4 until the accuracy is equal to 100%, and then the next step is performed.

[0103] A8: Sensitive parameter extraction

[0104] The significance parameter predicted by the random forest model in step A7 is extracted. When the time sequence significance deviation is not greater than ±3s, it is considered as a sign of damage, and the significance parameter is extracted as a monitoring sensitive parameter of subsequent damage.

[0105] The specific implementation is as follows:

[0106] B1, in the state of the thermal data collection and arrangement, the original bench working condition data record increment needs to meet the 1s recording interval, and the manual asynchronous recording data such as explosion pressure, supercharger speed and smoke degree are aligned according to the time label to form a relatively complete thermal parameter data resource; The data is arranged based on the fault shutdown time, and the thermal data within 2min before shutdown is extracted. If the 2min before shutdown does not include diesel engine normal working data, it is considered that the arrangement data does not meet the damage damage judgment criterion, and the test bench equipment can be warned and the data is excluded.

[0107] B2, in the data preprocessing process, unit transformation is carried out according to the data sample less than 1 criterion to prevent inaccurate identification of the training model. After the 2min data result of each set is sorted according to the time increment, the diesel engine thermal parameter has damping, and the order of each parameter is calculated IQR according to 7 points. The abnormal value is removed according to Q1-IQR and Q3+IQR as the reference, which does not cause abnormal judgment. After the data is removed, the missing result is linearly interpolated, and the expression is as follows:

[0108] ;

[0109] In the formula, is the data to be interpolated at t time, is the data at the last recording time, is the time that needs to be linearly interpolated, is the last time of the data to be interpolated at t time, is the next time of the data to be interpolated at t time;

[0110] Further, the date and time series are deleted.

[0111] B3, in the construction of moving window function, the window size N is initially defined as N=30, and the window function expression is as follows:

[0112] ;

[0113] where i is the window index (value range is 1 to 30);

[0114] The function expression of the target result in the window is as follows:

[0115] ;

[0116] where i is the window index, N is the window size (N = 30);

[0117] B4、In the feature value calculation, based on the data samples obtained in step B2, the change speed and change acceleration of each variable are extracted, and the step B2 result is combined into a data set, and further, the data set is divided by sliding segmentation according to the window size, and the samples of each variable within 2 minutes are taken at an interval of 1 sample in step B3 N samples, forming M new sequence matrix.

[0118] B5、In the data set division, first, the sequence matrix of each variable generated in step B4 is taken as training data, the window data constructed in step B3 is taken as target data, the total data set is synthesized, and the sample sequence is randomly shuffled, and the data set is divided according to the sample amount of 7:3, N×7 / 10 samples are taken as the training set, and N×3 / 10 samples are taken as the test set.

[0119] B6、In the random forest model training, the training set and test set obtained in step B5 are normalized using the following expression:

[0120] ;

[0121] where i is the window index, is the normalized feature, is the minimum value in the feature , is the maximum value in the feature ;

[0122] According to the number of 50 decision trees and 1 minimum leaf number, a random forest model is constructed, the training set is trained using the normalized result, and the test set is further predicted using the normalized result, and the accuracy of the model prediction result and the actual result is extracted.

[0123] ​B7, in the accuracy rate determination of the test set, the accuracy of the identification of the damage failure symptom recognition opportunity is focused on. Based on the accuracy rate of the model prediction result and the actual result extracted in step B6, if the accuracy rate is less than 100%, the window size N of the window function in step B3 is increased, and step B4 is restarted. The accuracy rate of the model prediction result and the actual result extracted in step B6 is 100%. After multiple iterations, the final N value is determined. In order to reduce the number of iterations, the 2 classification boundary point of the window function in step B3 is manually intervened to be placed in the normal time 10s before the diesel engine running damage.

[0124] B8, in the sensitive parameter extraction, the importance output result of the random forest model prediction accuracy of 100% is used to find the results of the adjacent importance indicators greater than 0.3 and the corresponding moving window time sequence, and the 2 classification boundary point of the B3 window function is determined as the best marker opportunity, and the full sample is input without sliding window processing, the training target of the whole sample before the best marker opportunity is 0, and the training target of the whole sample after the best opportunity is 1. The full flux random forest test is carried out, and the parameters and indicators with importance greater than 0.3 are extracted.

[0125] As shown in Figure 2 , a diesel engine bench destructive damage early warning method is as follows:

[0126] C1, diesel engine full working condition data preparation

[0127] It includes arranging a set of historical operating universal characteristic data of the target diesel engine.

[0128] C2, test data working condition classification

[0129] It includes dividing into 16 segmented working conditions according to the speed-load map.

[0130] C3: random forest model construction and training

[0131] It includes random model construction, training data set construction and output prediction model; in matlab, 24 sensitive thermal parameters are used to merge each parameter with the speed, torque and classification mark formed in step C2, and used as the classification mark training target for model training, and 24 trained model codes are output.

[0132] C4: test bench model deployment

[0133] It includes random forest prediction model deployment and working condition classification map query; in the real-time host of the test bench process control, 24 parallel loop random forest prediction models are arranged, and the running process working condition classification query map is arranged.

[0134] C5: Collecting warning set in diesel engine operation

[0135] Including warning code, used to form the control system to quickly identify the work logic.

[0136] C6: Determine whether the warning set is equal to the damage knowledge set

[0137] Including damage knowledge code, warning code and damage knowledge code comparison, used to determine whether there is an alarm condition.

[0138] C7: Warning and driving the emergency action position of the test bench

[0139] Including setting method, used to determine whether the predicted driving and test bench action logic are the same.

[0140] The specific implementation is as follows:

[0141] D1, in the diesel engine full working condition data preparation, a set of diesel engine historical operation universal characteristic data is sorted out, including the working condition change process and the stable working process, the data recording interval is 1s, and each working condition point includes the working condition value in the entering process and the stable 5min.

[0142] D2, in the test data working condition classification, according to the speed-load map, it is divided into 16 sub-working conditions, among which, in the vehicle characteristic, the maximum torque speed to the idle working speed is divided into independent working conditions, the maximum torque speed to the rated point speed range is divided into 5 speed gears, and the load is divided into three zones according to 0%-50%, 50%-75% and 75%-100%, and the speed is classified and identified from small to large-load from small to large; In the propulsion characteristic, the (highest speed-lowest speed) / 2 to the idle working speed is divided into an independent sub-working speed, and the other working condition classification is the same as the vehicle.

[0143] D3: In the random forest model construction and training, set the parameters, use each sensitive parameter of 24 thermal parameters combined with speed and torque as a data set, perform normalized data processing, merge with the classification mark formed in step D2, and take step D2 as the training target, monitor whether the error of each combined parameter after training reaches 100%, if the training accuracy is insufficient, adjust the classification range, and finally output the 24 random forest models after training as C code;

[0144] D4: In the random forest model deployment of the test bench, first perform normalization processing, and in real-time species construction, 24 random forest model codes are parallel round seeking, and at the round seeking time, the normalized data of the corresponding sensitive thermal parameters in the previous 7 are input to carry out prediction; Synchronously deploy the working condition classification query map of step D2 in the test bench; At the same time, compare the random forest prediction classification and the working condition query classification, if the prediction mark is the same as the running working condition classification mark, it is recorded as "0", if it is different, it is recorded as "1".

[0145] D5: Collecting the warning set in the diesel engine operation; sequentially sorting the 24 parameters in step D4 to form a 16 hexadecimal numerical code.

[0146] D6: In the determination of whether the warning set is equal to the damage knowledge set, the 24 thermal parameters are sorted according to the sorting order of the warning set collected in step D5, and are converted into knowledge codes, and the damaging damage annotation content is marked, when the warning set and the knowledge set coding result are the same, the warning indication is output, and the emergency setting symbol is pushed; if the warning set and the knowledge set coding result are different, step D5 warning set collection is continued;

[0147] D7: Warning and driving setting bench emergency action station, associate warning action word with bench soft emergency stop, execute process logic, if the bench is determined as emergency stop state with "1", the warning in step D6 is unchanged, if the bench is determined as emergency stop state with "0", the output result is reversed.

[0148] Example one:

[0149] E1, with the state of thermal data collection and arrangement, the data compliance identification of diesel engine damaging damage identification is carried out, which has 2min above slow parking process, including diesel engine cooling parking operation specification, diesel engine abnormal fault damage degree is not high, which does not meet the transient process, the original control process warning of test bench or diesel engine controller warning action, specific screening comparison as shown in Figures 3-4 .

[0150] E2, in the data preprocessing process, the problem of data quality is solved, including the parameter jump introduced by unstable test bench acquisition instrument, and the data acquisition label alignment and data interpolation of external mobile device under the insufficient test bench integration capability, the specific solution of data quality before and after the comparison as shown in Figures 5-6 .

[0151] E3, in the construction of moving window function, the problem of inaccurate detection time and subsequent use of parameters is solved, the moving window is constructed, the label time position is determined according to the classification accuracy, the window data is as shown in Figure 7 .

[0152] E4, in the calculation of characteristic value, the sliding window size is used to control the construction of new sequence under the combination of E2 and parameter change speed, acceleration data, the new sequence matrix is as shown in Figure 8 .

[0153] E5, in the data set division, the variable sequence matrix generated in step E4 is combined with the window data constructed in step E3, as shown in Figure 9The total data set is synthesized and randomly shuffled in sample sequence, and the data set is divided according to the sample size of 7:3, and the sample of N x 7 / 10 is used as the training set, and the sample of N x 3 / 10 is used as the test set.

[0154] E6, in the random forest model training, the effect before the sliding window size adjustment is as shown in Figure 10 .

[0155] E7, in the test set accuracy determination, the effect of the adjusted sliding window size is as shown in Figure 11 .

[0156] E8, in the sensitive parameter extraction, the importance output result of the random forest model prediction accuracy 100% is used to find the result with importance index greater than 0.3 and the sequence corresponding to the moving window moment, and the parameter is determined as no less than 2 as the criterion to determine the best marker opportunity of the 2 classification demarcation point of the window function in step E3, as shown in Figure 12 . Figure 13

[0157] After the random forest model recognition, the sensitive recognition parameter table corresponding to the destructive damage shown in the following table is as shown in Table 1:

[0158] Table 1

[0159]

[0160] Example two:

[0161] F1, in the diesel engine full working condition data preparation, the continuous recording working condition change process is adopted, as shown in Figure 14 .

[0162] F2, in the test data working condition classification, 16 segmented working conditions are divided according to the speed-load map, as shown in Figure 15 .

[0163] F3: in the random forest model construction and training, the 16 working condition classifications of each parameter are predicted as shown in Figure 16 .

[0164] F4: in the diesel engine operation, the sequence of each thermal parameter and warning code in the warning set is shown in Table 2.

[0165] Table 2

[0166] ID bit 24 23 22 21 20 19 18 17 Parameter RPM Output load Oil temperature Coolant pressure Main gallery pressure Boost pressure Diesel pressure Exhaust pressure ID bit 16 15 14 13 12 11 10 9 Parameter Exhaust pressure Water temperature difference Oil temperature difference Exhaust temperature Camshaft pressure Intake pressure Exhaust pressure rate Exhaust rate of change ID bit 8 7 6 5 4 3 2 1 Parameter Water temperature difference delta acceleration Oil pressure delta acceleration Exhaust temperature acceleration Camshaft pressure rate Main gallery pressure rate Diesel pressure rate RPM fluctuation rate Output load fluctuation rate

[0167] ​F5: In the determination of whether the warning set is equal to the damage knowledge set, the warning knowledge coding table is shown in Table 3.

[0168] Table 3

[0169] Serial number Damage event Status flag bit Warning code 1 Piston ring breakage 011000001000000100000000 608100 2 Cylinder liner damage 000100001100000010000000 10A080 3 Main bearing bush burn 010010000010000001000000 682040 4 Piston scuffing 000010010010000001000000 092040 5 Rocker arm damage 000010000001100000010000 081810 6 Connecting rod big end drop 010010000000100000001000 480808 7 Oil pump gear breakage 000000100001010000000100 021404 8 Intermediate idler breakage 100110000000000000000010 980002 9 Valve clatter 010000010000001000000001 610201

[0170] Advantages and beneficial effects of the present application:

[0171] (1) The present application provides a diesel engine damage monitoring and early warning method based on thermal data driving, which is used for rapid monitoring and early warning of diesel engine damage on the test bench, improves the safety of the diesel engine test bench, and solves the problem that in the traditional verification process, the operator with rich verification experience is used as the test master system, due to the fatigue factors of the continuous monitoring process, the differences in personnel perception and operation timeliness, and the high mechanical load of the main working parts of the diesel engine, the fault occurs in a short time, and it is difficult for the personnel to determine and respond effectively from monitoring experience to meet the rapid disposal of the test process.

[0172] (2) In order to advance the test bench early warning opportunity and accurately determine the sensitivity of the thermal parameter, avoid the influence of artificial experience on parameter determination, a sliding window function and a random forest model are constructed, the damage initiation detection opportunity marker position is evaluated according to the index, and the sensitive parameter under the corresponding detection opportunity is further determined, so as to avoid the problem that the light, high-strength and high-value diesel engine verification appears failure and is scrapped.

[0173] (3) In order to deploy the warning strategy in the test bench and have the damage warning pushing function, the present application provides a diesel engine bench damage early warning method, which codes the thermal parameter detection knowledge and the deviation amount, and reports the warning in real time through the test bench round monitoring mode, so that the subsequent expanded identification model can be continuously increased, and the test bench process control system is used.

[0174] (4) In order to solve the problem that more state abnormal data training is needed in the traditional process, the present application uses the deviation characteristics of normal working parameters, combines the known damage knowledge to realize the abnormal state warning process, the training prediction data source is relatively easy to obtain, the generated abnormal state warning process also has combination, avoids the misoperation introduced by single parameter abnormality, reduces the false alarm rate of the test bench, and provides a reliable method for reliable development of the test process.

[0175] (5) The application provides a diesel engine destructive damage monitoring and early warning method based on thermal data driving, solves the conversion of inventory thermal parameters to bench monitoring and early warning, reduces the demand of high risk on the working strength of test personnel, simultaneously has an alarm coding process, meets test bench process control deployment and positioning information pushing functions, makes the operator quickly discharge the abnormal state according to the alarm information, and decides the subsequent test progress, has a positive effect on diesel engine performance and reliability verification work. Meanwhile, based on the sliding window marking fault process, the prediction accuracy criterion can accurately grasp the abnormal state generation opportunity and sensitive parameters, reduces the artificial experience dependence, and the abnormal state identification knowledge has remarkable effect.

[0176] The above merely describes the preferred embodiments of the present application and is not intended to limit the present application. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A diesel engine destructive damage monitoring and early warning method based on thermal data driving, characterized in that: The method comprises the following steps: S1, capturing the sign of the diesel engine bench destructive damage and identifying sensitive parameters; S2, early warning of the diesel engine bench destructive damage; In step S1, capturing the sign of the diesel engine bench destructive damage and identifying sensitive parameters, comprises: S11, collecting and arranging thermal data to form a data resource; S12, based on the data resource, pre-processing the data to form a data sample; S13, based on the thermal data, constructing a moving window function for the reflection time of the destructive failure; S14, based on the data sample, calculating characteristic values to form a data sample matrix; S15, based on the moving window function and the data sample matrix, dividing the data set to form a training set and a test set; S16, based on the training set and the test set, training a random forest model and obtaining the prediction accuracy of the test set; S17, based on the prediction accuracy of the test set, determining the accuracy of the test set; S18, based on the random forest model, extracting sensitive parameters; In step S2, early warning of the diesel engine bench destructive damage, comprises: S21, preparing diesel engine full-condition data; S22, based on the diesel engine full-condition data, classifying test data conditions; S23, based on the test data conditions of step S22, constructing and training a random forest model to output a random forest prediction model; S24, based on the random forest prediction model of step S23, deploying a test bench model; S25, during diesel engine operation, collecting a warning set, wherein the warning set includes a warning code used to form a control system for quickly identifying working logic; S26, determining whether the warning set is equal to a destructive damage knowledge set, if yes, jumping to step S25, otherwise, proceeding to the next step, wherein the destructive damage knowledge set includes a destructive damage knowledge code, and the warning code is compared with the destructive damage knowledge code to determine whether there is an alarm condition; S27, alarming and driving the emergency action position of the test bench; In step S18, based on the random forest model, extracting sensitive parameters, comprises: Based on the random forest model, extracting a saliency parameter as a monitoring sensitive parameter for subsequent destructive damage; setting a time sequence saliency deviation as a sign of the destructive damage.

2. The diesel engine catastrophic damage monitoring and early warning method based on thermodynamic data driving according to claim 1, characterized in that: In step S11, collecting and arranging thermal data to form a data resource, comprises: Collecting historical stock fault process segment data, including test bench synchronous acquisition data and auxiliary burst pressure, supercharger speed and smoke data collected by mobile detection instruments; Based on the damage site combination when the fault occurs, classifying to form a data resource.

3. The diesel engine catastrophic damage monitoring and early warning method based on thermodynamic data driving according to claim 1, characterized in that: In step S12, based on the data resource, pre-processing the data to form a data sample, comprises: Performing unit transformation on the data resource, using the median value method to remove outliers, and using linear interpolation to supplement data for missing data items to form a data sample.

4. The diesel engine catastrophic damage monitoring and early warning method based on thermodynamic data driving according to claim 1, characterized in that: In step S13, based on the thermal data, constructing a moving window function for the reflection time of the destructive failure, comprises: Multiplying the reflection time of the thermal data on the destructive failure by a set multiple as the initial size of the window; Using the median of the window as the classification boundary to construct a classification function.

5. The thermodynamic data driven diesel damaging fault monitoring and early warning method according to claim 1, characterized in that: In step S14, based on the data sample, calculating characteristic values to form a data sample matrix, comprises: The data sample of step S12 is extracted by time increment to obtain the value change speed and value change acceleration, and is combined with the original data to form a data set; Each thermal parameter in the data set is divided by the construction window size; The division process is intercepted by setting the sliding window to form a data sample matrix.

6. The thermodynamic data driven diesel damaging fault monitoring and early warning method according to claim 1, characterized in that: In step S15, based on the moving window function and the data sample matrix, the data set is divided to form a training set and a test set, including: Based on the moving window function, the target training data is obtained, combined with the data sample matrix, the sample order sequence is shuffled, and the training set and the test set are divided according to the set mode.

7. The thermographic data-driven diesel engine catastrophic damage monitoring and alerting method of claim 1, wherein: In step S16, based on the training set and the test set, the random forest model is trained, and the test set prediction accuracy is obtained, including: The thermal parameter data in the training set and the test set is normalized; The normalized training set data is substituted into the random forest model, and the moving window function is trained as the training target; The test set is tested to obtain the test set prediction accuracy.

8. The thermographic data-driven diesel engine catastrophic damage monitoring and alerting method of claim 1, wherein: In step S17, based on the test set prediction accuracy, the test set accuracy is determined, including: Based on the test set prediction accuracy, the adaptability of the moving window function is determined; if the accuracy is less than the set value, the window size is increased, and the process is restarted at step S14 until the accuracy is equal to the set value.

9. The thermographic data-driven diesel engine catastrophic damage monitoring and alerting method of claim 1, wherein: In step S21, the diesel engine full working condition data is prepared, including: Organizing a set of diesel engine historical operation universal characteristic data; In step S22, the test data working condition classification, including: According to the form of speed-load map, a certain number of divided working conditions are divided; In step S23, the random forest model construction and training, including: Random model construction; training data set construction; output prediction model; In step S24, the test bench model deployment, including: Random forest prediction model deployment; Working condition classification map query; Arranging a certain number of parallel cycle random forest prediction models; Arrange the working condition classification query map of the running process.

Images