Engine abnormality diagnosis method, engine abnormality diagnosis program, and engine abnormality diagnosis system

Abstract

The engine abnormality diagnosis method, engine abnormality diagnosis program and engine abnormality diagnosis system of the present invention perform the following steps: step (S2): obtaining the initial state quantity of the engine model (10); step (S3): applying the initial state quantity and utilizing the engine model (10); step (S4): obtaining the estimated state quantity; step (S5): obtaining the measured state quantity; step (S6): applying the residual of the measured state quantity and the estimated state quantity to a nonlinear Kalman filter; step (S10): applying the Kalman gain to the engine model (10) and repeating steps (S4) to (S6); step (S11): calculating the correlation of the measured state quantity or the residual; step (S12): performing factor analysis to obtain the factor loading; step (S13): calculating the factor score and detecting the abnormality; step (S16): applying the factor loading to machine learning; step (S17): diagnosing the abnormality based on machine learning; and step (S18): outputting the diagnosis result.

Description

Technical Field

[0001] The present invention relates to an engine abnormality diagnosis method, an engine abnormality diagnosis program and an engine abnormality diagnosis system for diagnosing engine abnormalities using a mathematical engine model. Background Art

[0002] From the viewpoints of safety and efficiency, it is extremely important to detect abnormalities in engines mounted on ships and the like at an early stage.

[0003] Here, in Patent Document 1, a performance estimation system for a gas turbine engine is disclosed, which utilizes a constant gain extended Kalman filter (CGEKF) composed of a Kalman filter gain designed to be optimal at a certain operating point of the engine and a nonlinear dynamic simulation model that faithfully models the actual engine. When a difference occurs between an observed variable and an estimated value of the observed variable, the adjustment parameters of a part of the state variables constituting the model are adjusted to minimize the difference, thereby making the nonlinear dynamic simulation model a model that is always faithful to the actual engine. In addition, in paragraph 0010, an engine performance estimation system and method are described, which use a nonlinear dynamic simulation model of the engine and an estimation filter to always estimate, detect and monitor performance degradation caused by aging or damage caused by engine use.

[0004] In addition, Patent Document 2 discloses a system for determining flight parameters of an aircraft in real time during flight, the system being based on flight dynamics equations that determine the interdependence between at least two pre-selected flight parameters that should be estimated for the aircraft, and including an extended Kalman filter that is formed to provide a joint estimate of the selected flight parameters during flight of the aircraft.

[0005] In addition, Patent Document 3 discloses a Kalman filter configured to provide a new process variable estimate to a controller in each execution cycle of the controller so that the controller can generate a control signal for controlling the process, and to generate an estimate of the process variable value based on a slow or intermittent process feedback signal.

[0006] Prior art literature

[0007] Patent Literature

[0008] Patent Document 1: Japanese Patent Application Publication No. 2009-68359

[0009] Patent Document 2: Japanese Patent Application Publication No. 2013-49408

[0010] Patent Document 3: Japanese Patent Application Publication No. 2018-88289 Summary of the invention

[0011] Technical problem to be solved by the invention

[0012] In Patent Document 1, as described in paragraph 0005, "a high-precision nonlinear dynamic simulation model can be formed that fully reflects the effects of age-related degradation or damage that changes the correlation between measurable items and unmeasurable items." Patent Document 1 is an invention that sequentially tunes a nonlinear dynamic simulation model, estimates the unmeasurable items of the engine based on the measurable items, and uses the estimation results to control the gas turbine engine, rather than an invention that performs early detection of engine abnormalities and diagnosis of their causes.

[0013] Patent Document 2 is an invention for simultaneously estimating at least two flight parameters that are related to each other by flight dynamics equations by an extended Kalman filter and using them for flight control of an aircraft, but is not an invention for early detection of engine abnormalities and diagnosis of their causes.

[0014] Patent Document 3 is an invention related to the improvement of a Kalman filter in a process control system such as a chemical process or a petroleum process, and is not an invention for early detection of an abnormality in an engine and diagnosis of its cause.

[0015] Therefore, an object of the present invention is to provide an engine abnormality diagnosis method, an engine abnormality diagnosis program, and an engine abnormality diagnosis system capable of early detection of engine abnormality and diagnosis of the cause thereof.

[0016] Solutions for solving the above technical problems

[0017] The abnormality diagnosis method of the engine corresponding to the record of Scheme 1 is a method for diagnosing abnormalities of the engine using a mathematical engine model, characterized in that the following steps are performed: an initial state quantity acquisition step, acquiring the initial state quantity of the engine model; an engine model utilization step, applying the initial state quantity to the engine model and utilizing the engine model; an engine state estimation step, calculating the state of the engine based on the initial state quantity using the engine model and obtaining an estimated state quantity; a measured state quantity acquisition step, acquiring the measured state quantity of the engine; a Kalman filtering step, applying the residual between the acquired measured state quantity and the calculated estimated state quantity to a nonlinear Kalman filter; and repeating The invention discloses a method for detecting an abnormality in a plurality of engine states, comprising: applying the Kalman gain obtained by applying the nonlinear Kalman filter to the engine model, repeating the engine state estimation step, the measurement state quantity acquisition step, and the Kalman filtering step; a correlation calculation step, calculating the correlation of the measurement state quantity or the residual when applied to the nonlinear Kalman filter; a factor analysis step, performing factor analysis on the correlation of the measurement state quantity or the residual to obtain the factor loading; an anomaly detection step, calculating the factor score according to the factor loading and detecting the anomaly; a machine learning application step, applying the factor loading to the machine learning; an anomaly diagnosis step, diagnosing the anomaly based on the machine learning; and an output step, outputting the abnormality information including the diagnosis result of the abnormality of the engine.

[0018] According to the present invention described in Scheme 1, by using the measured state quantity or the residual difference between the measured state quantity and the calculated estimated state quantity to perform factor analysis, the abnormality of the engine can be detected early based on the calculated factor score. In addition, the factor loading can be applied to machine learning to diagnose the cause of the abnormality of the engine.

[0019] The present invention described in Scheme 2 is characterized in that a model updating step is further executed, and a result obtained in the Kalman filtering step or a result obtained by processing the acquired measured state quantity is applied to the engine model utilizing step to update the engine model.

[0020] According to the present invention described in claim 2, the engine model can be updated and the calculation accuracy of the estimated state quantity can be always maintained at a high state.

[0021] The present invention described in claim 3 is characterized in that the calculation of the correlation of the measured state quantity or the residual in the correlation calculation step is performed based on the correlation matrix.

[0022] According to the present invention described in claim 3, the calculation accuracy of the correlation of the measured state quantity or residual can be improved through simple calculation.

[0023] The present invention described in claim 4 is characterized in that, in the factor analysis step, singular value decomposition (SVD) is performed based on a covariance matrix as a correlation matrix to derive factor loadings.

[0024] According to the present invention described in claim 4, it is possible to extract essentially important contents without being restricted by the shape of the covariance matrix, and detect engine abnormality at an early stage.

[0025] The present invention described in claim 5 is characterized in that a self-organizing map (SOM) is used for machine learning.

[0026] According to the present invention described in claim 5, the cause of the abnormality of the engine can be classified using the self-organizing map (SOM) which is unsupervised machine learning.

[0027] The present invention described in claim 6 is characterized in that the initial state quantity acquired in the initial state quantity acquisition step is the engine load (Q p ) and includes the fuel pump rack position (h p ) of the fuel supply.

[0028] According to the present invention described in claim 6, it is possible to estimate the engine load (Q p ) and includes the fuel pump rack position (h p )’s fuel supply amount, and obtains the estimated state quantity.

[0029] The present invention described in claim 7 is characterized in that the measured state quantity acquired in the measured state quantity acquisition step is the engine speed (n e ).

[0030] According to the present invention described in claim 7, the engine speed (n) which is frequently measured can be e ) is used for engine abnormality diagnosis and can also improve measurement accuracy.

[0031] The present invention described in claim 8 is characterized in that the supercharger speed (n tc ), scavenging pressure (P s ), scavenging air temperature (T s ) and exhaust temperature (T e ) as the estimated state quantity in the engine state estimation step.

[0032] According to the present invention described in claim 8, it is possible to obtain the supercharger speed (n tc ), scavenging pressure (P s ), scavenging air temperature (T s ), or exhaust temperature (T e )’s diagnosis results.

[0033] The engine abnormality diagnosis program corresponding to the record of Scheme 9 is a program for diagnosing engine abnormalities using a mathematical engine model, and is characterized in that a computer is enabled to execute the initial state quantity acquisition step, engine model utilization step, engine state estimation step, measurement state quantity acquisition step, Kalman filtering step, repetition step, correlation calculation step, factor analysis step, abnormality detection step, machine learning application step, abnormality diagnosis step and output step in the engine abnormality diagnosis method.

[0034] According to the invention described in claim 9, by using the measured state quantity or the residual between the measured state quantity and the calculated estimated state quantity to perform factor analysis, the abnormality of the engine can be detected early based on the calculated factor score. In addition, the factor loading can be applied to machine learning to diagnose the cause of the abnormality of the engine.

[0035] The engine abnormality diagnosis system corresponding to the record of Scheme 10 is characterized in that it comprises: an engine; a condition input unit for inputting the initial state quantity of the engine model; a state quantity measurement unit for measuring the state of the engine and obtaining the measured state quantity; a computer for executing the engine abnormality diagnosis method or the engine abnormality diagnosis program; and an information providing unit for providing abnormality information output by the computer including the diagnosis result of the engine abnormality.

[0036] According to the present invention described in claim 10, it is possible to provide abnormality information including the result of early detection of an abnormality in the engine and diagnosis of its cause using a computer.

[0037] The present invention described in claim 11 is characterized in that the engine model is updated by a computer.

[0038] According to the present invention described in claim 11, the calculation accuracy of the estimated state quantity can be easily improved by updating the engine model using a computer.

[0039] The present invention described in claim 12 is characterized in that the engine speed (n e ) as the measured state quantity.

[0040] According to the present invention described in claim 12, the engine speed (n) which is frequently measured can be e ) is used for engine abnormality diagnosis and can also improve measurement accuracy.

[0041] The present invention described in claim 13 is characterized in that the information providing unit provides the supercharger speed (n tc ), scavenging pressure (P s ), scavenging air temperature (T s ) and exhaust temperature (Te ) is considered as an abnormal diagnostic result.

[0042] According to the present invention described in claim 13, it is possible to obtain the supercharger speed (n tc ), scavenging pressure (P s ), scavenging air temperature (T s ), or exhaust temperature (T e )’s diagnosis results.

[0043] The present invention described in claim 14 is characterized by comprising an abnormality control means for controlling the engine at the time of abnormality based on the output of abnormality information.

[0044] According to the present invention described in claim 14, by controlling the engine at the time of an abnormality, it is possible to prevent the abnormality of the engine from becoming serious or causing a failure.

[0045] The present invention described in claim 15 is characterized in that the abnormality information is provided using a human-machine interaction unit as the information providing unit.

[0046] According to the present invention described in claim 15, by providing engine abnormality information from the human-machine interaction unit, the crew and the like can respond quickly and appropriately when an abnormality occurs.

[0047] The present invention described in claim 16 is characterized in that it includes a sending unit that sends the abnormality information provided by the information providing unit to another place.

[0048] According to the present invention described in claim 16, it is possible to obtain the diagnosis result including the abnormality of the engine in real time even at a location far away from the engine.

[0049] The present invention described in claim 17 is characterized in that it includes a connection unit that connects the state quantity measuring unit, the computer, and the information providing unit online.

[0050] According to the present invention described in claim 17, it is possible to transmit and provide diagnosis results including engine abnormalities in real time online.

[0051] Effects of the Invention

[0052] According to the abnormality diagnosis program of the engine of the present invention, it is possible to perform factor analysis by using the measured state quantity or the residual of the measured state quantity and the calculated estimated state quantity, and detect the abnormality of the engine at an early stage based on the calculated factor score. In addition, the factor loading quantity can be applied to machine learning to diagnose the cause of the abnormality of the engine.

[0053] In addition, when the model updating step of applying the result obtained in the Kalman filtering step or the result obtained by processing the acquired measured state quantity to the engine model utilization step to update the engine model is further executed, the engine model can be updated so that the calculation accuracy of the estimated state quantity is always maintained at a high level.

[0054] Furthermore, when the calculation of the correlation of the residual in the correlation calculation step is performed based on the correlation matrix, the calculation accuracy of the correlation of the measurement state quantity or the residual can be improved through simple calculation.

[0055] Furthermore, in the factor analysis step, when singular value decomposition (SVD) is performed based on the covariance matrix as a correlation matrix to derive factor loadings, essentially important contents can be extracted without being restricted by the shape of the covariance matrix, thereby detecting engine abnormality at an early stage.

[0056] Furthermore, when a self-organizing map (SOM) is used as machine learning, the cause of the abnormality of the engine can be classified using the self-organizing map (SOM) which is unsupervised machine learning.

[0057] In addition, the initial state quantity acquired in the initial state quantity acquisition step is the engine load (Q p ) and includes the fuel pump rack position (h p ) of the fuel supply amount, it is possible to estimate the engine load (Q p ) and includes the fuel pump rack position (h p )’s fuel supply amount, and obtains the estimated state quantity.

[0058] In addition, the measured state quantity acquired in the measured state quantity acquisition step is the engine speed (n e ), the engine speed (n e ) is used for engine abnormality diagnosis and can also improve measurement accuracy.

[0059] In addition, when obtaining the supercharger speed (n tc ), scavenging pressure (P s ), scavenging air temperature (T s ) and exhaust temperature (T e ) as the estimated state quantity in the engine state estimation step, it is possible to obtain the supercharger speed (n tc ), scavenging pressure (P s ), scavenging air temperature (T s ), or exhaust temperature (T e )’s diagnosis results.

[0060] In addition, according to the abnormality diagnosis program of the engine of the present invention, it is possible to perform factor analysis by using the measured state quantity or the residual of the measured state quantity and the calculated estimated state quantity, and detect the abnormality of the engine at an early stage based on the calculated factor score. In addition, the factor loading quantity can be applied to machine learning to diagnose the cause of the abnormality of the engine.

[0061] Furthermore, according to the engine abnormality diagnosis system of the present invention, it is possible to provide abnormality information including the result of early detection of engine abnormality and diagnosis of its cause using a computer.

[0062] Furthermore, when the engine model is updated by a computer, the calculation accuracy of the estimated state quantity can be easily improved by updating the engine model by a computer.

[0063] In addition, the engine speed (n e ) as the measured state quantity, the engine speed (n e ) is used for engine abnormality diagnosis and can also improve measurement accuracy.

[0064] In addition, when the supercharger speed (n tc ), scavenging pressure (P s ), scavenging air temperature (T s ) and exhaust temperature (T e ) is used as the abnormality diagnosis result, it is possible to obtain the supercharger speed (n) useful for estimating the state of the engine. tc ), scavenging pressure (P s ), scavenging air temperature (T s ), or exhaust temperature (T e )’s diagnosis results.

[0065] Furthermore, in the case of providing an abnormality control unit that controls the engine in an abnormality based on the output of abnormality information, by controlling the engine in an abnormality, it is possible to prevent the abnormality of the engine from becoming serious or causing a failure.

[0066] Furthermore, when the abnormality information is provided using the human-machine interaction unit as the information providing unit, the abnormality information of the engine is provided from the human-machine interaction unit, so that the crew and the like can respond promptly and appropriately when an abnormality occurs.

[0067] Furthermore, when a transmission unit is provided for transmitting the abnormality information provided by the information providing unit to other places, a diagnosis result including an abnormality of the engine can be obtained in real time even at a place far from the engine.

[0068] Furthermore, when a connection unit is provided for connecting the state quantity measuring unit, the computer, and the information providing unit online, the diagnosis result including the abnormality of the engine can be transmitted and provided online in real time. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 This is a block diagram of an engine abnormality diagnosis system according to an embodiment of the present invention.

[0070] Figure 2 It is a flowchart of the abnormality diagnosis procedure of its engine.

[0071] Figure 3 This is a schematic diagram of the abnormality diagnosis of the engine.

[0072] Figure 4 This is a conceptual diagram of its factor analysis.

[0073] Figure 5 2 is a diagram showing an example of factor analysis using measurement data.

[0074] Figure 6 2 is a diagram showing examples of factor scores.

[0075] Figure 7 It is a graph showing the change in factor loading of each measurement value.

[0076] Figure 8 It is a diagram showing an example of a mathematical model of the engine.

[0077] Fig. 9 It is a diagram showing the parameters of the engine model.

[0078] Fig.10 This is a diagram showing the concept and calculation formula of the prediction (estimation) step and the update (correction) step of the Kalman filter.

[0079] Fig.11 : is a diagram showing the relationship between the Kalman filter and the factor analysis.

[0080] Fig.12 FIG. 2 is a diagram showing an example of abnormality detection using factor scores. DETAILED DESCRIPTION

[0081] Next, an engine abnormality diagnosis method, an engine abnormality diagnosis program, and an engine abnormality diagnosis system according to an embodiment of the present invention will be described.

[0082] Figure 1 This is a block diagram of an engine abnormality diagnosis system.

[0083] The engine abnormality diagnosis system uses a mathematical engine model 10 to diagnose abnormality of an engine 1 mounted on a ship or the like.

[0084] The abnormality diagnosis system of the engine comprises: a condition input unit 2 for inputting the initial state quantity of the engine model 10; a state quantity measuring unit 3 for measuring the state of the engine 1 and obtaining the measured state quantity; a computer 4 for executing the abnormality diagnosis method or program of the engine; an information providing unit (human-machine interaction unit: HMI) 5 for providing abnormality information including the diagnosis result of the abnormality of the engine 1 output by the computer 4; an abnormality control unit 6 for controlling the engine 1 in the abnormality based on the output of the abnormality information; a sending unit 7 for sending the abnormality information provided by the information providing unit 5 to other places; and a connecting unit 8 for connecting each device online. The engine model 10 is preferably corresponding to the engine 1 and installed in the computer 4 from the beginning, but when a part or all of the engine 1 is changed to a different state in association with the maintenance of the ship, for example, it can also be changed in the middle through the condition input unit 2 and the like.

[0085] The connection unit 8 is, for example, a router or a LAN, and connects the state quantity measurement unit 3, the computer 4, and the information providing unit 5 online. Thus, it is possible to transmit and provide the diagnosis result including the abnormality of the engine 1 online in real time. In addition, the online connection can be made by either wireless or wired via the connection unit 8.

[0086] The sending unit 7 is provided in the computer 4. In addition, the computer 4 includes a control unit 11, an initial state quantity acquisition unit 12, an engine state estimation unit 13, a measured state quantity acquisition unit 14, a Kalman filter unit 15, a repetition unit 16, a correlation calculation unit 17, a factor analysis unit 18, an anomaly detection unit 19, a machine learning application unit 20, an anomaly diagnosis unit 21, an engine model update unit 22, a main memory 23, an auxiliary memory 24, an engine model utilization unit 27, an output unit 28, and the like.

[0087] The auxiliary memory 24 is, for example, a hard disk. The auxiliary memory 24 stores an engine model 10, a factor load space 25, and a mechanical device failure database 26. The engine model 10 is a model pre-constructed based on the specifications and characteristics of the engine 1. In the factor load space 25, factor load data is accumulated after an abnormality of the engine 1 is detected. In addition, in the mechanical device failure database 26, data collected by simulating failures of the engine 1 through a simulation program are collected. In addition, in the case where there is an engine state value when a failure actually occurs, the engine state value can also be collected in the mechanical device failure database 26.

[0088] The condition input unit 2 is a mouse, a keyboard, a touch panel, etc. The operator or user 101 of the computer 4 uses the condition input unit 2 to input the initial state quantity. As the initial state quantity, the load (Q p ) and includes the fuel pump rack position (h p ) of the fuel supply amount. Thus, it is possible to estimate the load (Q) of the engine 1 which is important in estimating the state of the engine 1. p ) and includes the fuel pump rack position (h p ) to obtain the estimated state quantity. The input initial state quantity is sent to the initial state quantity acquisition unit 12 of the computer 4 and acquired. In addition, when the operating conditions change sufficiently to change the engine model 10, a condition signal can also be input through the condition input unit 2.

[0089] The state quantity measuring unit 3 is various sensors, etc. In the state quantity measuring unit 3, it is preferable to obtain the rotation speed (n e ) as the measured state quantity. Thus, the engine 1 speed (n e ) is used for abnormal diagnosis of the engine 1, and the measurement accuracy can be improved. In addition, as the measured state quantity measured by the state quantity measuring unit 3, there are other state quantities such as the scavenging pressure (P s ), exhaust temperature (T e ) and the load of engine 1 (Q p ) etc. The measured state quantity measured by the state quantity measuring unit 3 is sent to the measured state quantity acquiring unit 14 of the computer 4. In addition, as the measured state quantity, in addition to the measured value itself, it can also be a value obtained by processing based on the measured value, such as the average value of the measured value for a certain period or the difference from the average value.

[0090] The output unit 28 outputs the diagnosis result of the engine 1 obtained by the computer 4 to the transmission unit 7 , the information providing unit (human-machine interaction unit) 5 , and the abnormality control unit 6 .

[0091] In addition, the information providing unit 5 , the abnormality control unit 6 , and the transmitting unit 7 may also be provided outside the computer 4 .

[0092] The information providing unit 5 preferably provides the supercharger speed (n tc ), scavenging pressure (P s ), scavenging air temperature (T s ) and exhaust temperature (T e ) is used as the diagnosis result of the abnormality of the engine 1. Thus, it is possible to obtain the supercharger speed (n tc ), scavenging pressure (P s), scavenging air temperature (T s ), or exhaust temperature (T e ) diagnostic result. In addition, the information providing unit 5 can provide not only the diagnostic result of the abnormality of the engine 1, but also all the information related to the abnormality including the abnormality detection information.

[0093] The transmission unit 7 transmits the output diagnosis result to a device located at a different place from the computer 4 via a wired or wireless LAN via the connection unit 8. Thus, even at a place far from the engine 1, abnormality information including the diagnosis result of the abnormality of the engine 1 can be obtained in real time. Figure 1 In FIG. 1 , as a place to receive abnormal information, a bridge 100 and a user 101 on land such as a shipping company are shown. For example, when the computer 4 is in the engine room, the information providing unit 5 transmits the abnormal information to the bridge 100 via the in-vessel LAN.

[0094] The information providing unit (human-machine interaction unit) 5 is, for example, a monitor, a speaker, etc. By providing abnormal information of the engine 1 from the information providing unit (human-machine interaction unit) 5, the crew and the like can respond quickly and appropriately when an abnormality occurs. In addition, in addition to the human-machine interaction unit, the information providing unit 5 also includes all units related to the provision of information, such as a storage unit for temporarily storing abnormal information and a transmission unit for transmitting abnormal information to, for example, a smartphone.

[0095] The abnormality control unit 6 controls the engine 1 automatically or according to the operation from the crew when an abnormality occurs based on the abnormality detection result or the abnormality diagnosis result. By controlling the engine 1 when an abnormality occurs, it is possible to prevent the abnormality of the engine 1 from becoming serious or causing a failure. In addition, the information or signal provided to the abnormality control unit 6 may be the unprocessed information or signal provided to the information providing unit 5.

[0096] Figure 2 This is a flow chart of the engine abnormality diagnosis method. Figure 3 It is a schematic diagram of the abnormality diagnosis of the engine. In addition, since the abnormality diagnosis method of the engine can be provided as a program, the abnormality diagnosis program of the engine will be described below as a program that causes a computer to execute each step in the abnormality diagnosis method of the engine.

[0097] When the personal computer starts executing the engine abnormality diagnosis program, the control unit 11 reads the engine model 10 stored in the auxiliary memory 24 (engine model reading step S1).

[0098] After the engine model reading step S1 , the initial state quantity acquisition unit 12 acquires the initial state quantity of the engine model 10 input by the condition input unit 2 (initial state quantity acquisition step S2 ).

[0099] After the initial state quantity acquisition step S2, the control unit 11 applies the initial state quantity to the read engine model 10 in the engine model utilization unit 27 and utilizes the engine model 10 (engine model utilization step S3). In the engine model utilization step S3, the engine model 10 is initially set or updated. In order to use the engine model 10, it is necessary to apply any initial state quantity. As long as the initial state quantity (such as load or fuel amount) that is as close to the current state as possible is applied in the engine model utilization step S3 of utilizing the engine model 10, the engine model 10 can be utilized, such as reducing the calculation time in the engine model 10 and improving the accuracy of the estimated calculation performed based on the engine model 10. In addition, by utilizing the input of the residual in the Kalman filter step S6 in the engine model utilization step S3, or the result obtained by processing the acquired measured state quantity, the engine model 10 can be updated, and the engine model 10 can be utilized by faithfully reproducing the real engine 1.

[0100] The engine state estimation unit 13 calculates the state of the engine 1 based on the initial state quantity using the engine model 10 and obtains the estimated state quantity (engine state estimation step S4).

[0101] Furthermore, the measured state quantity acquisition unit 14 acquires the measured state quantity of the engine 1 obtained by the measurement performed by the state quantity measurement unit 3 (measured state quantity acquisition step S5 ).

[0102] The Kalman filter unit 15 inputs the residual between the acquired measured state quantity and the calculated estimated state quantity to the nonlinear Kalman filter (Kalman filter step S6). In the Kalman filter step S6, state estimation and model parameter estimation are performed.

[0103] After the Kalman filter step S6 , the engine model updating unit 22 determines whether the engine model 10 needs to be updated (model update determination step S7 ).

[0104] When the engine model updating unit 22 determines "yes" in the model updating determination step S7, that is, when the engine model 10 needs to be updated, the result obtained in the Kalman filtering step S6 or the result obtained by processing the acquired measured state quantity is applied to the engine model utilization step S3, and the engine model 10 is updated by updating the model parameters (model updating step S8). By updating the engine model 10, the calculation accuracy of the estimated state quantity can be always maintained at a high state corresponding to the aging degradation of the engine 1. In addition, by updating the engine model 10 through the computer 4, the calculation accuracy of the estimated state quantity can be easily improved. The determination in the model updating determination step S7 is performed by pre-setting a threshold value for the model parameters.

[0105] On the other hand, if the determination in the model update determination step S7 is "No", that is, if the engine model update unit 22 determines that the engine model 10 does not need to be updated, the repetition unit 16 determines whether a predetermined time has passed (time elapse determination step S9). The repetition unit 16 repeats the process k, k+1, k+2, etc., for example, in time divisions of 0.1 seconds.

[0106] When the repetition unit 16 determines as "No" in the time passage determination step S9, that is, when the specified time has not passed, the Kalman gain obtained by inputting into the nonlinear Kalman filter is applied to the engine model 10, and the engine state estimation step S4, the measurement state quantity acquisition step S5, and the Kalman filtering step S6 are repeated (repeat step S10).

[0107] Thus, the two functions of the Kalman filter, namely, state estimation and model parameter estimation, are used to repeat updating and state estimation of the engine model 10. Alternatively, when the measured state quantity is reliable, the Kalman filter can be used to identify the model parameters using a tracking filter and perform only state estimation.

[0108] Furthermore, the nonlinear Kalman filter is preferably an unscented Kalman filter or an extended Kalman filter, thereby making the Kalman gain more appropriate for the engine 1 as a nonlinear system and improving the calculation accuracy of the estimated state quantity.

[0109] On the other hand, when the determination result in the time elapse determination step S9 is "yes", that is, when the repeating unit 16 determines that the predetermined time has elapsed, the correlation calculation unit 17 calculates the correlation of the residual obtained when input to the nonlinear Kalman filter (correlation calculation step S11). The calculation of the correlation of the residual in the correlation calculation step S11 is preferably performed based on the correlation matrix. Thus, the calculation accuracy of the correlation of the residual can be improved by simple calculation.

[0110] In addition, the correlation can also be calculated using the acquired measurement state quantity instead of the residual obtained when input to the nonlinear Kalman filter.

[0111] After the correlation calculation step S11, the factor analysis unit 18 performs factor analysis on the correlation of the residuals to obtain factor loadings (factor analysis step S12).

[0112] In addition, it is also possible to perform factor analysis using the correlation of the acquired measurement state quantity instead of the correlation of the residuals to obtain the factor loading.

[0113] After the factor analysis step S12, the anomaly detection unit 19 calculates the factor score based on the factor loading amount and detects anomalies (anomaly detection step S13). The anomaly detected in the anomaly detection step S13 can also be output as anomaly information.

[0114] In the factor analysis step S12, singular value decomposition (SVD) is performed based on the covariance matrix as the correlation matrix to derive factor loadings. This allows essentially important contents to be extracted without being restricted by the shape of the covariance matrix, and allows abnormality of the engine 1 to be detected at an early stage.

[0115] The factor loadings obtained in the factor analysis step S12 are accumulated in the factor loading space 25 (factor loading accumulation step S14). The machine learning application unit 20 reads out the data stored in the mechanical device failure database 26 (mechanical device failure data reading step S15), and applies the factor loadings to the machine learning (machine learning application step S16). As machine learning, a self-organizing map (SOM) is used. Thus, the cause of the abnormality of the engine 1 can be clearly classified using the SOM as unsupervised machine learning. In addition, as a machine learning algorithm, in addition to SOM, SVM (Support vector machine), fuzzy C-means, etc. can also be used.

[0116] The abnormality diagnosis unit 21 diagnoses abnormality based on machine learning (abnormality diagnosis step S17 ).

[0117] After the abnormality detection step S13 and the abnormality diagnosis step S17, the output unit 28 outputs abnormality information including the abnormality diagnosis result of the engine 1 (output step S18). In the output of the abnormality information in the output step S18, in addition to the abnormality diagnosis result, the abnormality detection information and the accompanying information can also be included. The output destination is the information providing unit (human-machine interaction unit) 5, the abnormality control unit 6 and the sending unit 7 as described above.

[0118] The abnormality diagnosis of the engine 1 using factor analysis will be described in detail below.

[0119] Figure 4 This is a conceptual diagram of factor analysis. Factor analysis is to explore the m common, its relation is determined by a im The analysis of the hidden (unmeasured) variable factor F is represented by the linear relationship parameter a as shown in the following formula (1).

[0120] [Number 1]

[0121]

[0122] In formula (1), f is a common factor, a im is the linear coefficient, and the remaining u i It is measurement error or noise caused by unexplained factors.

[0123] Each measurement value y m Linearly connected to a common factor f of a certain quantity. Linear coefficient a im This is called factor loading. Factor analysis is about exploring hidden variables (factors), but factors can also be considered as anomalies (cause of accidents) when measured values ​​change.

[0124] Figure 5 is a diagram showing an example of factor analysis using measurement data. Figure 5 As shown in the figure, if the calculation example of the main factor and factor loading A based on the specific tracking change of the measured value is taken as an example of factor analysis using the measured data Y, the measured values ​​y1, y2, ... m For example, to represent the scavenging pressure, exhaust temperature, and engine load of the engine 1, a certain time interval (for example, the interval from 0 to t) is aggregated to create a matrix ( Figure 5 Calculate the covariance matrix of this matrix and divide it by the standard deviation of each variable to find R( Figure 5 (b)). Perform singular value decomposition (SVD) on R to find the singular value S. Next, find the first factor loading A ( Figure 5 (c)). The value obtained by adding the squares of the first factor loadings a and dividing by the sum of the variances is the factor score D (index) ( Figure 5 (d)).

[0125] Furthermore, it is possible to obtain the factor loadings at the first stage by performing singular value decomposition (SVD) on the correlation matrix of the covariance matrix, and to further obtain characteristic and significant factor loadings by using the EM (Expectation Maximization) method.

[0126] Figure 6 is a graph showing an example of factor scores, Figure 6 (a) shows the original factor score F1(D1), Figure 6 (b) shows the filtered factor score F1(D1). In addition, Figure 7 is a graph showing the changes in factor loadings for each measurement value. Figure 7 (a) is the scavenging pressure, Figure 7 (b) is the supercharger speed, Figure 7 (c) is the exhaust temperature, Figure 7 (d) is the scavenging temperature. If the factor scores D1 (primary index) are arranged in time series, for example, Figure 6 As shown, changes can be seen when an abnormality (some kind of change) occurs in the engine 1, such as when the flow rate of the cooling water of the air cooler of the engine 1 starts to decrease.

[0127] In other words, if Figure 7 As shown, the information contained in the factor loading matrix A represents an anomaly caused by a certain change in the propulsion system. Furthermore, the factor loading (row of the matrix A) represents the strength of the relationship between the propulsion system parameters as the elements of the error, and represents the characteristics of the cause of the anomaly. Therefore, by performing machine learning on the factor loading using a machine learning algorithm, such as a self-organizing map (SOM), the factor loading can be used to classify the causes of the propulsion system accident. In addition, the factor score D1 can be used for early detection of anomalies.

[0128] like Figure 3 As shown, in the present invention, a Kalman filter observer is used as the Kalman filter unit 15, and here, the residual between the measured state quantity and the estimated state quantity is used for factor analysis. That is, in the present invention, as another method, the residual E between the measured state quantity and the estimated state quantity obtained by the engine state estimation unit 13, which is inherently calculated by the Kalman filter unit 15, is used instead of the measured value (measured state quantity) Y, and the same factor analysis as above is performed. The residual E refers to the deviation from the estimated state quantity, and if the estimated state quantity is considered to be in a normal state, it reflects the deviation from the normal state. Therefore, based on the residual E, it is possible to detect that a certain abnormality has occurred in the engine 1.

[0129] The Kalman filter observer is based on the digital twin engine model 10. In the Kalman filter observer, the residual E of the measured state quantity obtained through the dynamic process of the engine 1 and the estimated state quantity calculated through the process of the mathematical engine model 10 based on the initial state quantity is input into the nonlinear Kalman filter. The Kalman gain is obtained. The Kalman gain is applied to the engine model 10 for process control of the mathematical engine model 10.

[0130] In this way, by monitoring the engine state using the engine model 10 as a digital twin model of the engine 1 , it is possible to detect a failure of the engine 1 at an early stage and diagnose the cause.

[0131] Figure 8 is a diagram showing an example of a mathematical model of an engine.

[0132] Figure 8 The left side of the arrow in the figure is the actual fuel supply system in the engine 1 (the fuel pump rack position (h p )) and the speed of engine 1 (n e ) etc., showing the measurement points and measurement values ​​in the engine 1. Figure 8 The right side of FIG. 2 shows an engine model 10 as a mathematical model of the engine 1 .

[0133] In addition, Figure 8The letters enclosed by ○ in the figure indicate the state quantity measuring unit 3. "T" enclosed by ○ is a thermometer, "P" enclosed by ○ is a pressure gauge, "n" enclosed by ○ is a tachometer, and "Q" enclosed by ○ is a shaft horsepower meter. The load change is measured by the shaft horsepower meter.

[0134] The measured value is measured for each step (k, k+1, k+2, ...), and the engine model 10 is used to perform calculation (estimation) for each step. In order to improve the estimation accuracy of the engine model 10, the speed (n) of the engine 1 that can be measured most reliably and accurately is obtained. e ) to calculate the Kalman gain and gradually correct the engine model 10. This is the Kalman filter.

[0135] exist Figure 8 In the mathematical model of the engine 1 shown, the behavior under normal conditions can be represented by a nonlinear state space model. The nonlinear state space model is composed of an equation represented by a state equation X, various parameters represented by state quantities x, and inputs represented by input quantities u. The right side of the state equation X is represented by various functions represented by the box in the center of the right side, and the state quantities n e 、n tc , P s , T e , P e , G f With input h p , Q p The relationship is expressed in five formulas.

[0136] For example, take the fuel pump rack position (h p ) and the engine load (Q p ) as input, take the speed of engine 1 (n e ), turbocharger speed (n tc ), scavenging pressure (P s )、Exhaust pressure(P e ), exhaust temperature (T e ), fuel flow rate (G f ) as the state (x) and output (y), representing the output function H of each system function F and the observation equation y(t).

[0137] exist Figure 8 In the example, the output observation value y is equal to the state quantity x.

[0138] Fig. 9 is a diagram showing parameters of the engine model.

[0139] The model parameters include engine loads such as propeller torque (Q p ), engine torque (Q e ), moment of inertia (Ie ,I tc ), fuel pump rack position (h p ), engine speed (n e ), turbocharger speed (n tc ), atmospheric pressure (P a ), atmospheric temperature (T a ), scavenging pressure (P s ), scavenging air temperature (T s ), maximum compression pressure in the cylinder (P c ), maximum combustion pressure in the cylinder (P z ), mean effective pressure in cylinder (P i ), scavenging air box volume (V a.r )、Exhaust box volume (V e.r ), thermodynamic constant (R a , R e , k e , C pe , C pa ), cooling water temperature (T w ), compressor outlet temperature (T c )、Exhaust pressure(P e ), exhaust temperature (T e ), turbine outlet temperature (T out ), turbine outlet pressure (P out ), fuel flow rate (G f ), scavenging air flow (G a )、Compressor air flow(G c )、Exhaust flow rate (G e ).

[0140] Fig.10 It is a diagram showing the concept and calculation formula of the prediction (estimation) step (Kalman filter step S6) and the update (correction) step (model update step S8) of the Kalman filter.

[0141] If the difference between the engine model 10 and the entity and the uncertainty of the measurement by the state quantity measurement unit 3 are taken into account, an error (residual) will occur between the estimated state quantity obtained based on the engine model 10 and the measured state quantity. The error is calculated, corrected by measuring the state quantity, and the estimated state quantity is made as close to the correct value as possible. This is the unscented Kalman filter as a nonlinear Kalman filter. Fig.10 The basis of the unscented Kalman filter is shown in .

[0142] By applying the Kalman filter, the uncertainty of measurement and modeling is taken into account at each discrete sampling time k, and the behavior of the propulsion system is repeatedly represented by the calculated error (the following equation (2)).

[0143] [Number 2]

[0144]

[0145] In the normal state of operation, the distribution of errors is considered to be a normal distribution with a mean of 0. In order to detect abnormal conditions, the errors generated by the Kalman filter are provided to factor analysis.

[0146] Fig.11 is a diagram showing the relationship between the Kalman filter and factor analysis.

[0147] The parameter A of the factor analysis model (Y=AF+U) is estimated from the correlation matrix (covariance matrix). As described above, the Kalman filter observer is based on the digital twin engine model 10, performs covariance estimation each time, and uses the Kalman gain as an indicator of which state quantity should be corrected.

[0148] Fig.12 FIG. 1 is a diagram showing an example of abnormality detection using factor scores.

[0149] Fig.12 This is an example of quickly detecting an abnormality in the engine abnormality diagnosis system by conducting a blockage simulation experiment of a supercharger suction filter of an actual engine 1. In addition, the horizontal axis of the graph is the number of samplings, and 1 time is 0.1 seconds. Although the supercharger suction filter gradually blocks and the pressure loss increases, the factor score F1 rises sharply at the initial stage when the pressure loss starts to increase, so the abnormality can be detected.

[0150] In this way, by performing factor analysis using the residual of the measured state quantity and the calculated estimated state quantity, it is possible to detect the abnormality of the engine 1 at an early stage based on the calculated factor score. In addition, the factor loading amount can be applied to machine learning to diagnose the cause of the abnormality of the engine 1. In addition, the computer 4 can provide abnormality information including the result of early detection of the abnormality of the engine 1 and diagnosis of the cause thereof.

[0151] In addition, the components and peripheral units of the computer 4 can be appropriately externally installed or built-in, and the functions of the computer 4 can be shared by a plurality of computers, or a part of them can be set as discrete circuits.

[0152] Industrial Applicability

[0153] The present invention can, for example, detect and diagnose an abnormality of an engine of a launched ship at an early stage, and can obtain the diagnosis result including the abnormality of the engine in real time even in a remote location, thus contributing to safe and efficient navigation. In addition, the present invention can also be used for the early detection and abnormality diagnosis of an abnormality of an engine other than a ship.

[0154] Description of Reference Numerals

[0155] 1 Engine

[0156] 2Conditional input unit

[0157] 3. State quantity measurement unit

[0158] 4. Computer

[0159] 5Information provision unit (human-computer interaction unit)

[0160] 6 Abnormal control unit

[0161] 7Sending unit

[0162] 8 connection units

[0163] 10 Engine Models

[0164] S2 initial state quantity acquisition steps

[0165] Steps to use the S3 engine model

[0166] S4 Engine state estimation step

[0167] S5 Measurement state quantity acquisition steps

[0168] S6 Kalman filter steps

[0169] S8 Model Update Steps

[0170] S10 repeat step

[0171] S11 related calculation steps

[0172] S12 Factor Analysis Steps

[0173] S13 Abnormality Detection Steps

[0174] S16 Machine Learning Application Steps

[0175] S17 abnormality diagnosis steps

[0176] S18 output step.

Claims

1. A method for diagnosing an abnormality of an engine, which is a method for diagnosing an abnormality of an engine using a mathematical engine model, characterized in that: Perform the following steps: An initial state quantity acquisition step of acquiring initial state quantities of the engine model constructed in advance based on the specifications and characteristics of the engine; an engine model utilizing step of applying the initial state quantity to the engine model and utilizing the engine model; An engine state estimation step, using the engine model to calculate the state of the engine based on the initial state quantity and obtain an estimated state quantity; a measurement state quantity acquisition step of acquiring a measurement state quantity of the engine that is a basis for constructing the engine model; A Kalman filtering step, applying the residual between the acquired measured state quantity and the calculated estimated state quantity to a nonlinear Kalman filter; Repeating the step of applying the Kalman gain obtained by applying to the nonlinear Kalman filter to the engine model, repeating the engine state estimation step, the measurement state quantity acquisition step, and the Kalman filtering step; A correlation calculation step, calculating a correlation matrix of the residuals of the measured state quantity and the estimated state quantity when applied to the nonlinear Kalman filter; A factor analysis step, performing factor analysis on the calculation results of the correlation matrix to obtain factor loadings; anomaly detection step, calculating factor scores according to the factor loadings, and detecting the anomalies; A machine learning application step, applying the factor loadings to machine learning; an abnormality diagnosis step, diagnosing the abnormality based on the machine learning; An output step of outputting a detection result of the abnormality of the engine and a diagnosis result of the abnormality.

2. The engine abnormality diagnosis method according to claim 1, characterized in that: A model updating step is further performed, and the result obtained in the Kalman filtering step, or the result obtained by processing the acquired measured state quantity as the average value of a certain period of the measured value or the difference from the average value is applied to the engine model utilization step to update the engine model.

3. The engine abnormality diagnosis method according to claim 1, characterized in that: In the factor analysis step, singular value decomposition (SVD) is performed based on the covariance matrix as the correlation matrix to derive the factor loadings.

4. The engine abnormality diagnosis method according to claim 1, characterized in that: The machine learning uses a self-organizing map or SOM.

5. The engine abnormality diagnosis method according to claim 1, characterized in that: The initial state quantity acquired in the initial state quantity acquisition step is the load (Q p ) and includes the fuel pump rack position (h p ) of the fuel supply.

6. The engine abnormality diagnosis method according to claim 1, characterized in that: The measured state quantity obtained in the measured state quantity obtaining step is the speed of the engine (n e ).

7. The engine abnormality diagnosis method according to claim 1, characterized in that: Get the supercharger speed (n tc ), scavenging pressure (P s ), scavenging air temperature (T s ) and exhaust temperature (T e ) as the estimated state quantity of the engine state estimation step.

8. A computer-readable storage medium storing an abnormality diagnosis program for an engine, the storage medium storing a program for abnormality diagnosis of an engine using a mathematical engine model, characterized in that: Cause a computer to execute the initial state quantity acquisition step, the engine model utilization step, the engine state estimation step, the measurement state quantity acquisition step, the Kalman filtering step, the repetition step, the correlation calculation step, the factor analysis step, the anomaly detection step, the machine learning application step, the anomaly diagnosis step and the output step in the engine abnormality diagnosis method described in any one of claims 1 to 7.

9. An engine abnormality diagnosis system, characterized in that: have: engine; A condition input unit, for inputting the initial state quantity of the engine model; A state quantity measuring unit, which measures the state of the engine and obtains a measured state quantity; A computer for executing the engine abnormality diagnosis method according to any one of claims 1 to 7 or the engine abnormality diagnosis program according to claim 8; The information providing unit provides abnormality information including a diagnosis result of an abnormality of the engine output by the computer.

10. The engine abnormality diagnosis system according to claim 9, characterized in that: The engine model is updated by the computer.

11. The engine abnormality diagnosis system according to claim 9, characterized in that: The speed of the engine (n e ) as the measured state quantity.

12. The engine abnormality diagnosis system according to claim 9, characterized in that: The information providing unit provides the supercharger speed (n tc ), scavenging pressure (P s ), scavenging air temperature (T s ) and exhaust temperature (T e ) as the diagnostic result of the abnormality.

13. The engine abnormality diagnosis system according to claim 9, characterized in that: An abnormality control unit is provided for performing control to prevent the abnormality of the engine from becoming serious or causing a failure based on the output of the abnormality information.

14. The engine abnormality diagnosis system according to claim 9, characterized in that: A human-computer interaction unit is used as the information providing unit to provide the abnormal information.

15. The engine abnormality diagnosis system according to claim 9, characterized in that: A sending unit is provided to send the abnormal information provided by the information providing unit to other places.

16. The engine abnormality diagnosis system according to claim 9, characterized in that: A connection unit is provided to connect the state quantity measuring unit, the computer and the information providing unit online.

Images