KR102204199B1 - Prediction method and diagonoisis method of vehicle for autonomous public bus - Google Patents

Abstract

The present invention relates to a method for performing vehicle fault diagnosis and prediction in real time with respect to an autonomous public bus (preferably, an electric bus). The method includes a process for determining whether vehicle components are normal and diagnosing a state of the vehicle components by collecting and analyzing data of vehicle sensors installed on the autonomous public bus so that the data is transmitted through a communication network in real time. Moreover, the method includes a framework for improving the diagnosis and prediction accuracy of the state of the vehicle components through learning with respect to basis data of the state diagnosis and state prediction of the vehicle components after predicting a change in the state of the vehicle components caused by maintenance activities to determine the maintenance activities in accordance with maintenance costs and performing a post-analysis therefor.

Description

Vehicle failure diagnosis and prediction method for autonomous public buses{PREDICTION METHOD AND DIAGONOISIS METHOD OF VEHICLE FOR AUTONOMOUS PUBLIC BUS}

The present invention relates to a monitoring system for an autonomous vehicle, and in particular, to a method for diagnosing and predicting failures for an autonomous public bus.

Various researches and developments are being carried out on autonomous vehicles. Prior Patent Document 1 discloses a method for controlling an autonomous vehicle. When an autonomous vehicle operates entirely dependent on information received from an external server, it cannot actively respond to various unexpected situations that may occur in the driving environment, so the autonomous vehicle is actively using information detected from sensors installed in the vehicle. Prior Patent Document 1 proposes a method that can be coped with. Whether the autonomous vehicle actively responds to the situation or the server controls the operation of the autonomous vehicle, technologies related to autonomous driving are mostly focused on research on passenger cars and vans.

However, public buses can also run autonomously, and in this case, safety and reliability are bound to become more important. Nevertheless, according to the prior art, there was no optimized solution for such an autonomous public bus.

The inventors of the present invention have completed the present invention after long research and development to implement a method and system for real-time monitoring of stability and reliability optimized for an autonomous public bus.

Prior Patent Document 1: Republic of Korea Patent Publication No. 10-2019-0078105

It is an object of the present invention to present a methodology for diagnosing and predicting failures for autonomous public buses, presupposing that the control system can grasp the reliability of an autonomous public bus in real time.

The term <self-driving public bus> refers to a public bus that runs autonomously without a driver. It is preferably an electric vehicle bus using a battery. In another embodiment, it may be a bus driven by an internal combustion engine.

The present invention communicates with an autonomous public bus in real time, and continuously collects data for measuring the state of vehicle parts through vehicle sensor data. In addition, the present invention performs vehicle failure diagnosis and prediction by grasping the characteristics of the collected vehicle part data. In addition, the present invention intends to implement a vehicle condition-based diagnosis prediction system capable of predicting vehicle component performance changes and costs according to maintenance activities based on this.

Since autonomous public buses are not managed by relying on drivers, the system must manage them in an integrated manner while communicating with all autonomous public buses. In addition, if a component breaks down or a battery problem occurs, the risk of an accident on a public bus that is driving increases, so an accurate framework and precise mathematical methodology are required.

On the other hand, other objects not specified of the present invention will be additionally considered within a range that can be easily inferred from the detailed description below and its effects.

In order to achieve the above object, the first aspect of the present invention is a vehicle failure diagnosis method for an autonomous public bus:

(a) The autonomous driving public bus monitoring server communicates in real time with N (N is an integer greater than 1) autonomous driving public buses to collect raw data from vehicle sensors of each autonomous driving public bus,

(b) the data management unit generates vehicle sensor data by purifying the vehicle sensor raw data, determining a range and unit of the data, and performing data normalization processing,

(c) The condition diagnosis unit collects the vehicle sensor data for a certain period to form a group of data to be analyzed, then detects an abnormality in parts by comparing it with a predetermined normal performance range or a normal performance cluster distribution range, and detects a plurality of pre-registered data groups. After diagnosing the part status using the status grade pattern data, after generating the part status diagnosis data,

(d) the autonomous driving public bus monitoring server transmitting the parts condition diagnosis data to the autonomous driving public bus management server and individually notifying the autonomous driving public bus.

A second aspect of the present invention is a vehicle failure diagnosis and prediction method for an autonomous public bus:

(a) The autonomous driving public bus monitoring server communicates in real time with N (N is an integer greater than 1) autonomous driving public buses to collect raw data from vehicle sensors of each autonomous driving public bus,

(b) the data management unit generates vehicle sensor data by purifying the vehicle sensor raw data, determining a range and unit of the data, and performing data normalization processing,

(c) The condition diagnosis unit collects the vehicle sensor data for a certain period to form a group of data to be analyzed, then detects an abnormality in parts by comparing it with a predetermined normal performance range or a normal performance cluster distribution range, and detects a plurality of pre-registered data groups. Diagnose the condition of the part using the condition grade pattern data, and generate the part condition diagnosis data,

(d) The state prediction unit derives a state change transition probability matrix based on the vehicle part state change occurrence frequency information, and generates part state prediction information that predicts the probability and timing of transition from the current part state to another state,

and (e) determining, by the maintenance determination unit, an optimal maintenance activity by calculating a minimum expected cost for maintenance activities based on the current state of the vehicle parts based on the parts condition prediction information.

The present invention can accurately monitor the operation stability and reliability of an autonomous public bus. The advantage is that it is possible to optimally inspect and manage autonomous public buses according to the condition of parts through real-time monitoring of various parts of autonomous public buses. For autonomous public buses to actually operate, the safety and reliability of the vehicle itself must be ensured very accurately. Therefore, accurate diagnosis and prediction of the condition of parts of a vehicle is as important as real-time management of driving. In that respect, the remarkable advantages of the present invention are highlighted.

On the other hand, even if it is an effect not explicitly mentioned herein, it is added that the effect described in the following specification and its provisional effect expected by the technical features of the present invention are treated as described in the specification of the present invention.

1 shows a schematic network system configuration example according to an embodiment of the present invention.
2 shows an example of the configuration of the server 100 according to a preferred embodiment of the present invention.
3 schematically shows a framework according to a preferred embodiment of the present invention.
4 shows the principle of data-based anomaly detection composed of a single element in a condition diagnosis process among a preferred embodiment of the present invention.
5 shows the principle of data-based anomaly detection composed of complex elements in a condition diagnosis process among a preferred embodiment of the present invention.
6 is a schematic diagram of a process of deriving a state transition probability matrix in a state prediction process among a preferred embodiment of the present invention.
7 shows a deterioration curve derived from a deterioration model inference process in a state prediction process among a preferred embodiment of the present invention.
8 is a diagram conceptually showing a process of determining an optimal maintenance activity in a maintenance determination process among a preferred embodiment of the present invention.
9 is a table showing the configuration of experimental data for battery components in the experimental example of the present invention.
10 is a diagram showing a condition diagnosis experiment in an experimental example of the present invention. See Figure 4.
11 is a table showing a pattern matching similarity analysis result for condition diagnosis in an experimental example of the present invention.
12 is a table showing results of verifying the accuracy of condition diagnosis in an experimental example of the present invention.
13 shows that the state transition probability matrix is derived in the state prediction experiment in the experimental example of the present invention. See Figure 6.
14 experimentally shows that the battery state trend can be predicted through deterioration model inference in the experimental example of the present invention. See Figure 7.
15 is a table showing the optimal maintenance activity at the present time during the maintenance determination experiment in the experiment of the present invention by priority.
※ The accompanying drawings are exemplified as reference for understanding the technical idea of the present invention, and the scope of the present invention is not limited thereto.

The present invention includes a process of collecting and analyzing vehicle sensor data of an autonomous public bus (preferably an electric vehicle bus) in real time to determine whether a vehicle component is normal and to diagnose a vehicle component state. In addition, after executing such a process, it predicts changes in the condition of vehicle parts due to maintenance activities, determines maintenance activities according to maintenance costs, performs post-analysis, and then analyzes the condition of vehicle parts and predicts the condition (status It may include a framework for improving the accuracy of diagnosis and prediction of vehicle parts status through learning about transition probability and pattern information for each status grade. Details of the vehicle failure diagnosis and prediction system to which the state-based maintenance technique of the present invention is applied will be described below.

The configuration of the present invention guided by various embodiments of the present invention and effects resulting from the configuration will be described with reference to the drawings. In describing the present invention, when it is determined that the subject matter of the present invention may be unnecessarily obscured as matters obvious to those skilled in the art with respect to known functions related to the present invention, a detailed description thereof will be omitted.

1 schematically shows an example of a system configuration according to a preferred embodiment of the present invention.

The autonomous driving public bus monitoring server 10 is installed on the autonomous driving public bus 1 while communicating in real time with N (N is an integer greater than 1) autonomous driving public bus 1 belonging to the network system of the present invention. Receives raw vehicle sensor data measured by various sensors (not shown).

Preferably, the autonomous public buses 1 without a driver are preferably electric vehicles driven by batteries rather than internal combustion engines.

The autonomous driving public bus monitoring server 10 may be composed of one or more server devices, and these server devices are composed of one or more hardware/software devices. A plurality of processors, storage, communication modems, signal processing devices, input/output devices, etc. may be included. In addition, it will include a user interface for managing each processor constituting the system of the present invention, a data search and inquiry support module.

In particular, in the present invention, for example, through the autonomous public bus (APB) monitoring system manager 100, a process such as diagnosing the condition of parts of an autonomous public bus, predicting the condition and determining maintenance, etc. I can.

In the system of the present invention, a plurality of databases will be built. The database 161 records and stores data collected in real time from a vehicle sensor. The database 163 includes parts status analysis target data, parts status rating pattern data, status frequency and status transition probability matrix data, which are data processed at each stage, expected maintenance cost by determining maintenance activities, and conditions before and after maintenance activities. It stores status history data such as comparison data. The database 165 may store information on each processor parameter, sensor information and threshold values, and data necessary for system operation.

On the other hand, the autonomous driving public bus monitoring server 10 of the present invention provides parts condition diagnosis data obtained through the vehicle failure diagnosis process described below, or parts condition prediction information obtained through the vehicle failure diagnosis and prediction process, and/or optimal Maintenance activity information may be transmitted to an autonomous public bus management server 200, which is a separate system server. In addition, the above data and information can be delivered through a situation board indicating the status of autonomous public buses in real time, the manager terminal of the subject managing autonomous public buses, web homepage or mobile. The self-driving public bus management server 200 may be composed of one or more server devices, and performs a task of integrated management of an autonomously-driving public bus in operation or waiting.

2 shows a schematic configuration example of an autonomous public bus monitoring server according to a preferred embodiment of the present invention.

First, the autonomous public bus monitoring system manager 100 includes, for example, a data management unit 110, a condition diagnosis unit 120, a condition prediction unit 130, a maintenance decision unit 140, and a post analysis unit 150. It can be configured.

The data storage 160 stores and manages data including a plurality of databases 161, 163, and 165 as described above. Preferably, the Hadoop system is used. It processes events such as storage/modification/retrieval of vehicle sensor information collected from autonomous public buses and information managed in each step of the process of the present invention system.

The distributed messaging system 170 facilitates real-time processing of various large amounts of data in the monitoring system of the present invention, and also serves to improve performance. It is used for linkage between internal processors and for storing and retrieving databases.

The V2X server 180 collects sensor information of N autonomous public buses in real time through a wireless network (eg, LTE and WAVE communication method). It can also deliver warning and notification information to each of these autonomous public buses. Preferably, SAE J2735, which is an autonomous vehicle V2X communication standard protocol, may be used. In addition, the data received from the autonomous public bus is decoded and transmitted to the distributed messaging system.

Although described again below, the system tasks executed by the modules included in the autonomous public bus monitoring system manager 100 will be briefly described as follows. The data management unit 110 performs data purification operations such as correction of missing values, data smoothing, and data normalization for the collected vehicle sensor information. The condition diagnosis unit 120 diagnoses the condition of a part through pattern analysis with data for each part condition grade and a part abnormality diagnosis by deviating from the normal performance range for the vehicle part analysis target information or through cluster distribution analysis. The state prediction unit 130 predicts the state of a part by a part state transition probability matrix according to maintenance activities (maintenance, repair, and replacement) using a data mining method. The maintenance determination unit 140 determines the optimal maintenance activity by calculating the expected cost of transition between states according to maintenance activities (maintenance, repair, replacement). The post-analysis unit 150 learns about the part state grade pattern information of the condition diagnosis unit 120 and the part state transition probability matrix of the state prediction unit 130 by using the part status result information before and after the final determined maintenance activity. Execute the process.

3 schematically shows a framework according to a preferred embodiment of the present invention implemented by the systems and modules described above.

Preferably, the framework of the present invention can be largely composed of five steps: data management (S10), condition diagnosis (S20), condition prediction (S30), maintenance decision (S40), and post-analysis (S50). In some preferred embodiments of the present invention, the component condition diagnosis data generated through the steps S10 and S20 may be transmitted to an external system, for example, an autonomous public bus management server, or may be individually notified to an individual autonomous public bus. In another preferred embodiment of the present invention, prediction information may be accumulated while performing all the procedures of steps S10 to S50, and the accumulated prediction information may be transmitted to an autonomous public bus management server.

First, look at the S10 data management process. The data management step may include data collection and monitoring (S11), data purification (S12), and data normalization (S13).

The raw sensor data collected in the data collection and monitoring process (S11) is data that can measure the status of vehicle parts such as speed, acceleration, vibration, load, voltage, and temperature. Vehicles that can continuously obtain data through real-time monitoring Collect for parts. As a process of acquiring raw data that can be used in the data purification process, a continuous data collection process that can observe changes in vehicle parts performance over time and further quantify performance is essential. Meanwhile, characteristics of vehicle parts and characteristics of collected data are registered in the system in advance.

Even if an advanced analysis method is applied for data analysis, if data that has not been data purified is used for analysis, meaningless results may be derived, so the data purification process (S12) is performed. In step S12, contradictory data is found, and missing value correction and data smoothing are performed to correct data inconsistency. Missing value correction is a process of minimizing loss and distortion of information by correcting missing values caused by malfunction of the observation sensor and the limit of the observation range. For example, the missing value correction can be performed as shown in Equation (1).

(1)

(One)

In Equation (1), A _t is the estimated value using the trend at the point t where the missing value occurs, n is the period for calculating the estimated value using the trend, and A _tn is the data collected at the point tn from the point where the missing value occurred.

Through data smoothing, when there are undesired irregular fluctuations due to noise during data extraction, such fluctuations or discontinuities are weakened or eliminated to smooth them.

(2)

(2)

In equation (2), A _t is the data collected in the current period, and A _t-1 is the data collected in the previous period t-1 .

(3)

(3)

In Equation (3), A _m is the weighted average of the current and previous period data, and At is the data collected at the t period. ω is the weight.

As shown in Equation (2), if the data collected in the current period is not within a certain range (±50%) based on the data of the previous period, as shown in Equation (3), the data of the previous period and the current period are collected. The resulting data is smoothed with a moving average by weight.

In the data normalization process, when condition diagnosis and prediction are performed by fusion of two or more different sensor data, analysis is difficult if the units and ranges of the data are different. Therefore, in this case, data normalization is performed with one unit and range. The normalization range at this time is substituted with a value between -1 and 1.

Next, the state diagnosis process of S20 will be described.

The condition diagnosis step is a step to grasp the current state of vehicle parts through data analysis. Preferably, it may include a data-based abnormality detection (S21) and a condition diagnosis process (S22).

In the data-based abnormality detection process (S21), when the diagnosis target data is composed of a single element, a method of distinguishing data outside the normal performance range of time series data may be used (see FIG. 4). In addition, when the data to be diagnosed is composed of complex elements, a method of discriminating data outside the normal performance cluster distribution range may be used (see FIG. 5).

For data-based anomaly detection consisting of a single element, the value of the normal performance range of vehicle parts is predetermined and registered in the system. And, as shown in Fig. 4, when the data deviates from the range of this value, it is determined as an anomalous value.

For data-based anomaly detection composed of complex elements, the normal performance range value of each vehicle part is predetermined and registered in the system, and then, as shown in FIG. 5, after grasping the normal performance cluster at the same coordinates, the distance d1 between the new data and the cluster Or, if the d2 value is outside the normal performance distribution range of C1 and C2, it is judged as an anomalous value. The normal performance range value is defined in the system according to the performance acceptance criteria set by the manufacturer for each vehicle part.

Next, the state diagnosis process (S22) will be described. The state level pattern data is determined for each predetermined state and registered in the system. Similarity analysis is performed through pattern matching between the diagnosis target data and each state level pattern data. The state level pattern data is data adjusted and learned using the state change history data before and after the past state diagnosis and maintenance activities. In this process, the similarity analysis through data-based pattern matching calculates the similarity value of the diagnosis target data and the condition grade pattern data for each condition using Equation (4), and among them, the condition grade pattern data whose similarity value is the smallest. The state can be determined as the final state of the data to be diagnosed.

(4)

(4)

In Equation (4), d(V _s , V _p ) means the similarity value between the diagnosis target data and the condition grade pattern data. V _s is the status level pattern data, V _p is the data to be diagnosed, V _{s, n} are the nth data of the status level pattern data, and V _{p, n} are the nth data of the data to be diagnosed.

Next, the state prediction process of S30 of the present invention will be described. This step may include a step of deriving a state transition probability matrix using a Markov process model (S31), and a step (S32) of inferring a deterioration model of a vehicle part according to a maintenance activity.

The step S31 is as illustrated in FIG. 6. It is a process of calculating a state transition frequency matrix from state change history data to create a state transition frequency matrix, and calculating the ratio of each element to the sum of occurrence frequencies for each row to derive a state transition probability matrix, which is schematically illustrated in FIG. Indicated. S _n displayed on the left and top of the matrix (a) and (b) of FIG. 6 denotes the status of vehicle parts. It shows that the frequency of state transition from S1 to S2 is 12, and dividing it by 51, the total frequency of transition from S1 to other states S1 to S5, gives a probability of 0.24 (24%).

In the deterioration model inference process in step S32, the performance change over time of future vehicle parts is predicted. In addition, in order to keep the state of the vehicle parts in a usable state, a deterioration model is derived through probabilistic inference of the performance degradation of the vehicle parts.

The deterioration model can be expressed as a deterioration curve as shown in FIG. 7. The horizontal axis represents the time period, the vertical axis represents the state of the vehicle part, t represents the current point in time, t +1 represents the minimum cycle unit for performing the condition check, and t+n represents the last life time of the vehicle component. The deterioration curve decreases with the passage of time due to deterioration of the parts, and the condition grade rises to s2 due to vehicle parts repair at the time t+a , and increases to s1 , the best condition due to the replacement of vehicle parts at the time t+b. Showing what to do.

The deterioration model in the state prediction stage is created through probabilistic inference using the state transition probability matrix as shown in FIG. 6 according to maintenance activities for vehicle parts and time information for the vehicle part state to transition to another state.

The deterioration model may be used when calculating the expected cost according to maintenance activities in the maintenance determination step of S40, which is the next step of the system framework. Now let's look at the maintenance decision process of the S40.

The maintenance determination step consists of a maintenance cost calculation process (S41) and a maintenance determination process (S42) according to maintenance activities.

8 and Equation (5) can be used to calculate the minimum expected cost for maintenance activities. 8 conceptually illustrates the maintenance optimization process. In this conceptual diagram, the status of vehicle parts is divided into grades 1 to 5, and non-response, repair, and replacement are indicated as 1, 2, and 3 as maintenance activities, respectively, and the status from the current point in time (Time t) to the planned point in time (T). By calculating the minimum expected cost of vehicle components at each inspection point, the optimal maintenance activities to maintain the performance of vehicle components are determined. The operation used to determine the optimal maintenance activity of FIG. 8 can use Equation (5). From the end of the planned period, the calculation is performed in a way that traces back the minimum expected cost for each maintenance activity.

(5)

(5)

In equation (5), V(i, t) is the minimum expected cost from the point t to the planned period T when the state is i , a is the maintenance activity (1=not responded, 2=repair, 3=exchange) , r is the discount rate, i is the state of the vehicle part at the present time, j is the state of the vehicle part at the next time, P _a (i, j) is the probability that the vehicle part will transition from state i to j by maintenance activity a , C (a, i) refers to the cost of action when performing maintenance activity a in state i .

The maintenance decision process (S42) derives results such as the minimum expected cost until the planned target period according to the maintenance activity and the priority of maintenance activities, and provides information to the system operator to make a maintenance decision.

Finally, the post-analysis step (S50) of the present invention includes a step (S51) of tracking changes in the state of vehicle parts before and after the maintenance activity, and a step (S52) of adjusting the algorithm-related data of the state diagnosis and prediction step. I can.

In the state tracking process (S51), after the system operator's decision on maintenance for vehicle parts is made and the maintenance response is completed, changes in the state of the vehicle parts before and after the maintenance activities for the vehicle parts are tracked and Create state change history data.

In the adjustment process of step S52, the state level pattern data used in the state diagnosis step, the state transition probability matrix data derived from the state prediction step, the deterioration model inference data, etc. are adjusted and learned to diagnose the state of the vehicle condition-based diagnosis and prediction system. The prediction accuracy is gradually increased to improve system reliability.

[Experimental example of the present invention]

The inventors of the present invention confirmed the feasibility of applying the framework and methodology of the vehicle condition-based diagnostic prediction system of the present invention as described above through experiments. Assuming that the condition of the battery part undergoes a probabilistic deterioration process by acquiring the data operated only by the battery engine of the hybrid electric vehicle, the experiment was conducted in stages of condition diagnosis, condition prediction, and maintenance decision.

The experimental data consists of 3,293 records, which are data on battery parts composed of 20 battery blocks as the vehicle power source. As shown in the table of FIG. 9, the collected battery component-related data consists of a battery charge amount, a battery current, and a voltage of each of the 20 battery blocks.

As for the deterioration process of battery components, it is assumed that the voltage difference between the battery blocks appears smaller as the new battery becomes, and the voltage difference between the battery blocks appears larger as the aging progresses. The difference between the maximum and minimum values of the battery block voltage was defined as dV. After calculating the dV value for each record of the experimental data and configuring it as time series data, it was used in the step of diagnosing the condition of battery parts, predicting the condition, and determining the maintenance.

(1) Condition diagnosis experiment

By substituting the experimental data consisting of dV values defined as the difference between the maximum and minimum values of the battery block voltage, diagnosis was performed for each predefined condition, and the accuracy was measured. The battery component status grade was classified into 5 grades using dV values. In the state of maintaining the normal performance, dV values are classified into four states with a dV value of 0.45V or less, 0.70V or less, 0.95V or less, and 1.2V or less, and an abnormal state is 3,293 dV values by dividing into one state above 1.2V. As a result, pattern data for each status level was organized.

Since dV values are time series data of a single element, battery component abnormality detection was performed based on whether or not it deviated from the range of the allowable performance reference value. As shown in FIG. 10, the dV performance tolerance criterion value was defined as 1.2V, and battery component abnormality detection was performed. As a result, it was found that 524 out of a total of 3,293 exceeded the performance tolerance criterion. The dV value is a value in which the dV value exceeds 1.2V for a certain period in order to judge it as a battery component because a spark may occur temporarily depending on the vehicle operation conditions such as vehicle start, accelerator and brake operation, etc. When this is maintained, it is necessary to judge it as a part abnormality.

In the area a of FIG. 10, dV values were 1.3V ~ 1.9V for 6 cycles. Since then, it was maintained at the normal performance standard, so it was not judged as a battery component abnormality. However, in the area b, the dV value was less than 1.2V for 772 cycles, and in some cases, it was mostly maintained at 1.3V ~ 3.5V, so in this case, it was judged as a battery component abnormality. Except for the area b, the range of change in battery current of the experimental data is maintained in the range of -50.0A to 20.0A, whereas in the area b, the range of change in battery current is in the range of -80.0A to 150.0A.

Next, the condition diagnosis was performed through data-based pattern analysis using the data to be diagnosed and the pattern data for each condition grade. In order to replace the new data collected in real time, the data to be diagnosed was assumed to be new data by selecting 20 data each in a time series from the experimental data.The pattern data for each state level was randomly extracted by 100 data and using Equation (4). By analyzing the pattern similarity, the state class calculated as the minimum similarity value was diagnosed as the state class to which the pattern data belongs.

The table of FIG. 11 is a result of analyzing the similarity of pattern matching as described above for condition diagnosis. A total of 165 data to be diagnosed were diagnosed as 53 cases for state 1, 52 cases for state 2, 23 cases for state 3, 14 cases for state 4, and 23 cases for state 5.

In order to verify the accuracy of the condition diagnosis, data to be diagnosed were selected one by one for each condition grade among the diagnosis subject data for which condition diagnosis was completed, and pattern data for each condition grade was randomly extracted to perform data-based pattern matching. This operation was repeated 100 times for each condition grade, and the result was the same as the table in FIG. 12. In the case of condition 1, 100 out of 100 cases were correctly diagnosed, in the case of condition 2, 96 cases, 100 cases in condition 3, 92 cases in condition 4, and 100 cases in condition 5 were correctly diagnosed. However, in the case of the data to be diagnosed with condition 2, one case was diagnosed as condition 1, three cases were diagnosed as condition 3, and eight cases were diagnosed as condition 5 in the case of condition 4 data to be diagnosed. For incorrectly diagnosing the normal state as an abnormal state, it is necessary to supplement the algorithm used in the experimental examples of the present invention, but it was confirmed that the result was significant as a whole.

(2) state prediction experiment

In order to predict the future state through deterioration model inference for battery parts, it is necessary to obtain historical data of state change before and after due to continuous long-term data collection, part condition diagnosis and maintenance activities (non-response, repair, exchange). It is possible.

Here, the state transition probability matrix as shown in FIG. 13 was derived using the state transition results before and after the state diagnosis time point of the 165 diagnosis target data used in the state diagnosis process.

In other words, if there is a state transition by comparing the state of the battery parts before and after the state of the data to be diagnosed based on the time of diagnosis, the state transition frequency matrix is calculated by adding 1 to the corresponding element, and the state in each row unit By calculating the ratio of each element to the total frequency of transition occurrence, the state transition probability matrix according to the unresponsive maintenance activity as shown in Fig. 13(a) was derived. On the other hand, since it is impossible to derive the state transition probability matrix according to the repair and replacement maintenance activities from the experimental data, it was arbitrarily produced as shown in (b) and (c).

By knowing the state transition probability matrix according to the non-correspondence, repair, and replacement maintenance activities of FIG. 13 and the performance maintenance time for each state, it is possible to predict the state of the battery through probabilistic deterioration model inference. The deterioration curve as shown in FIG. 14 was inferred by assuming the performance maintenance time of each state as t+4, t+3, t+2, and t+1, respectively, from 1 to 4 grade. If the planned target period is t+22 , when the state is S2 at the point of time t+7 from the current point (t), it is transferred to S1 to the repair and maintenance activity, and when the state is S4 after t+12 , the exchange is maintained. It indicates that the state is S1 due to the maintenance activity, and there is no maintenance activity since then, and it is in the state of S5 at the time t+22 .

(3) Maintenance decision experiment

For the maintenance decision experiment, the planned period, discount rate, maintenance activity cost by condition grade, and residual value by condition grade during the plan subject period are determined. In addition, by substituting the state transition probability matrix according to the maintenance activity derived from the previous state prediction experiment into Equation (5) and tracking back from the planned target period to the present time, the optimal maintenance activity at the present time is determined by priority in Fig. I got it like a table.

Here, plan target period (T) = 40 cycles, discount rate = 0.05, maintenance activity cost by status level (stage response = 0, repair for status 1-5 = 0.5, 8.5, 14.5, 55.5, 58.5, exchange = 60), The residual value (state 1~5 = -0.5, -128, -230, -300, -354) for each state level during the planned period is arbitrarily determined, and the state transition probability matrix according to the maintenance activity is executed by Equation (5). Each term in the matrix was finely adjusted with random values. This is the result of assuming that an adjustment process was made through tracking changes in the condition of vehicle parts before and after the decision on maintenance activities. The information constituting the table of FIG. 15 includes Current State = current state level, Minimum State = lowest allowed state, Action = optimal maintenance activity at the present time (1 = non-response, 2 = repair, 3 = exchange), LCC = Life cycle cost according to maintenance activity, Cost = cost according to maintenance activity.

(4) Experiment result

The vehicle condition-based diagnosis and prediction system of the present invention is required to secure the status change history data before and after due to continuous long-term data collection for vehicle parts and part condition diagnosis and maintenance activities (non-response, repair, exchange). The accuracy of diagnosis, prediction and maintenance decisions can be guaranteed. Therefore, as data accumulates, accurate diagnosis, prediction, and maintenance decisions will be made. This is an outstanding advantage of the present invention. Meanwhile, in the experimental example of the present invention described above, some data for the experiment are arbitrarily defined and used for the experiment due to the absence of state change history data before and after the maintenance activity and the data collected over a long period of time. Despite these limitations, through experiments, the framework and methodology for condition-based diagnosis and prediction presented in this study confirmed the possibility of diagnosing the condition of battery parts, inferring a probabilistic deterioration model, and determining maintenance activities. In that it did, a significant effect could be confirmed.

For reference, the vehicle failure diagnosis method and prediction method for an autonomous public bus according to an embodiment of the present invention may be implemented in the form of program instructions that can be executed through various computer means and recorded in a computer-readable medium. . The computer-readable medium may include program instructions, data files, data structures, and the like alone or in combination. The program instructions recorded on the medium may be specially designed and configured for the present invention, or may be known and usable to those skilled in computer software.

Examples of computer-readable media include hard disks, magnetic media such as floppy disks and magnetic tapes, optical recording media such as CD-ROMs and DVDs, magnetic-optical media such as floptical disks, and ROM, RAM, A hardware device specially configured to store and execute program instructions such as flash memory or the like may be included. Examples of program instructions include not only machine language codes created by a compiler, but also high-level language codes that can be executed by a computer using an interpreter or the like. The above-described hardware device may be configured to operate as one or more software modules to perform the operation of the present invention, and vice versa.

The scope of protection of the present invention is not limited to the description and expression of the embodiments explicitly described above. In addition, it is added once again that the scope of protection of the present invention may not be limited due to obvious changes or substitutions in the technical field to which the present invention pertains.

Claims (3)

(a) The autonomous public bus monitoring server communicates in real time with N (N is an integer greater than 1) autonomous public buses to collect raw data from vehicle sensors of each autonomous public bus,
(b) the data management unit generates vehicle sensor data by purifying the vehicle sensor raw data, determining a range and unit of the data, and performing data normalization processing,
(c) The condition diagnosis unit collects the vehicle sensor data for a certain period to form a group of data to be analyzed, then detects an abnormality in parts by comparing it with a predetermined normal performance range or a normal performance cluster distribution range, and detects a plurality of previously registered data groups. Diagnose the condition of the part using the condition grade pattern data, and generate the part condition diagnosis data,
(d) The state prediction unit derives a state change transition probability matrix based on the vehicle part state change occurrence frequency information, and generates part state prediction information that predicts the probability and timing of transition from the current part state to another state,
(e) Autonomous public bus comprising the step of determining, by a maintenance decision unit, an optimal maintenance activity by calculating a minimum expected cost for maintenance activities based on the current state of vehicle parts based on the parts condition prediction information For vehicle failure diagnosis and prediction method.
delete delete

Images