WO2022019020A1 - 異常検知装置及び異常検知プログラム - Google Patents
異常検知装置及び異常検知プログラム Download PDFInfo
- Publication number
- WO2022019020A1 WO2022019020A1 PCT/JP2021/023245 JP2021023245W WO2022019020A1 WO 2022019020 A1 WO2022019020 A1 WO 2022019020A1 JP 2021023245 W JP2021023245 W JP 2021023245W WO 2022019020 A1 WO2022019020 A1 WO 2022019020A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- time
- abnormality
- time window
- detection device
- series data
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Ceased
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
- B60W50/02—Ensuring safety in case of control system failures, e.g. by diagnosing, circumventing or fixing failures
- B60W50/0205—Diagnosing or detecting failures; Failure detection models
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M17/00—Testing of vehicles
- G01M17/007—Wheeled or endless-tracked vehicles
-
- G—PHYSICS
- G05—CONTROLLING; REGULATING
- G05B—CONTROL OR REGULATING SYSTEMS IN GENERAL; FUNCTIONAL ELEMENTS OF SUCH SYSTEMS; MONITORING OR TESTING ARRANGEMENTS FOR SUCH SYSTEMS OR ELEMENTS
- G05B23/00—Testing or monitoring of control systems or parts thereof
- G05B23/02—Electric testing or monitoring
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
- G06N20/10—Machine learning using kernel methods, e.g. support vector machines [SVM]
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
- B60W2050/0001—Details of the control system
- B60W2050/0043—Signal treatments, identification of variables or parameters, parameter estimation or state estimation
- B60W2050/005—Sampling
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
- B60W50/02—Ensuring safety in case of control system failures, e.g. by diagnosing, circumventing or fixing failures
- B60W50/0205—Diagnosing or detecting failures; Failure detection models
- B60W2050/021—Means for detecting failure or malfunction
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W2556/00—Input parameters relating to data
- B60W2556/10—Historical data
Definitions
- the present invention relates to a technique for detecting an abnormality in a vehicle signal.
- Patent Documents 1 and 2 reference data indicating normality is collected in advance using another vehicle at the time of the durability test of the vehicle, and the data measured during the durability test using the reference data. The technology to detect anomalies is described.
- the present invention has been made in view of the above points, and provides an anomaly detection device and an anomaly detection program capable of performing anomaly detection while suppressing the effects of individual differences in vehicles and wear during testing. Make it an issue.
- the abnormality detection device of the present invention includes a time window processing unit that extracts data included in the time window by dividing a plurality of vehicle signals including time series data by a time window.
- the learning unit that generates learned information by learning using the time-series data of the vehicle signal separated by the time window, and the time-series data of the vehicle signal separated by the time window are used as the learned information. It is characterized in that it is reconstructed based on the above, the error before and after the reconstruction is calculated, and the abnormality degree calculation unit for calculating the abnormality degree based on the calculated error and the learned information is provided.
- the abnormality detection program of the present invention is characterized in that the computer functions as the abnormality detection device.
- the present invention when detecting an abnormality in a vehicle, it is possible to suppress the influence of individual differences in the vehicle and wear during the test.
- the abnormality detection system 1 is a system for detecting an abnormality in the vehicle 2, and includes a measuring device 10, an abnormality detecting device 20, a notification device 30, and the like. To prepare for.
- the measuring device 10 measures various signals (vehicle signals) of the vehicle 2 during a simulated test run in the tabletop durability test, and transmits a plurality of measured vehicle signals to the abnormality detecting device 20.
- the measured vehicle signal includes signals of different orders and / or units, and includes wheel rotation speed (wheel speed), engine rotation speed, engine rotation torque, engine throttle opening, and engine temperature (exhaust temperature). , Water temperature, oil temperature), output torque of the drive motor, current generated by the drive motor, failure code from the engine control unit, and the like.
- the abnormality detection device 20 acquires a plurality of vehicle signals measured by the measuring device 10, and detects an abnormality for each of the acquired vehicle signals.
- the abnormality detection device 20 is composed of, for example, a CPU (Central Processing Unit), a ROM (Read-Only Memory), a RAM (Random Access Memory), an input / output circuit, and the like.
- the abnormality detection device 20 includes a sampling rate adjustment unit 21, a time window processing unit 22, an abnormality degree calculation unit 23, an abnormality determination unit 24, and a learning unit 25 as functional units.
- the sampling rate adjusting unit 21 acquires a plurality of (n types) vehicle signals measured by the measuring device 10 and adjusts and matches the sampling rates of the acquired plurality of vehicle signals (step S2 in FIG. 2).
- the plurality of vehicle signals include time series data for different sampling rates.
- the signal included in each vehicle signal is indicated by a black circle on the axis indicating the time.
- the sampling rate adjusting unit 21 is based on a predetermined sampling rate stored in advance, and among the data in the vehicle signal, the data closest to each of the timings (every time interval t) of the predetermined sampling rate (dotted circles in FIG. 4).
- the data enclosed by the mark is extracted as the data at the relevant timing, and the other data is thinned out.
- the sampling rate adjustment unit 21 outputs the vehicle signal whose sampling rate has been adjusted to the time window processing unit 22.
- the predetermined sampling rate is at least the highest sampling rate among the sampling rates of the plurality of vehicle signals.
- the sampling rate adjusting unit 21 can improve the correlation of each vehicle signal by such processing. Further, the sampling rate adjusting unit 21 can prevent overfitting of a vehicle signal having a high sampling rate.
- the time window processing unit 22 is mounted on the volatile memory 20a.
- the time window processing unit 22 acquires n types of vehicle signals whose sampling rates have been adjusted by the sampling rate adjustment unit 21, and from the vehicle signals whose sampling rates have been adjusted, a predetermined number (m pieces) included in the time window.
- Data B n is extracted (step S3 in FIG. 2).
- t is the slide width of the time window and the cycle of abnormality determination (time required to execute steps S2 to S9).
- B n (or B n (0) ) is the data (m ⁇ 1 matrix) included in the time window of this time.
- the time window processing unit 22 is a process of extracting m pieces of data included in the time window by sliding the time window for a predetermined period with the time t as the slide width. Generates B n.
- the slide width t of the time window may be smaller than the predetermined period of the time window. Further, it is desirable that the slide width t of the time window is the same as the time interval t of the predetermined sampling rate.
- the time window processing unit 22 outputs the generated B n to the abnormality degree calculation unit 23, and stores C n , which is a matrix in which B n is arranged, in the storage 20 b as storage information.
- C n is the storage information is represented by the following formula.
- B n (p) is data (m ⁇ 1 matrix) included in the time window p times before B n.
- C n is stored information (m ⁇ p matrix) stored in the storage 20b.
- p is the number of times the data Bn is extracted (in the present embodiment, a value equal to the number of times of abnormality determination in the abnormality determination unit 24), and increases as time passes from the start of operation of the abnormality detection device 20.
- the time window processing unit 22 enables real-time learning and abnormality calculation by dividing the vehicle signal into each input number m of the self-encoder 23a (see FIG. 7) described later.
- the abnormality degree calculation unit 23 acquires the learned information stored in the volatile memory 20a (step S1 in FIG. 2), and also acquires the B n generated by the time window processing unit 22. Then, the degree of abnormality of the vehicle signal is calculated based on the acquired Bn and the learned information.
- the learned information will be described in detail later.
- the abnormality degree calculation unit 23 includes the current B n (0) generated by the time window processing unit 22 and the previous C n, that is, B n (1) , B n (2) , ..., B.
- the degree of abnormality of the vehicle information is determined. calculate.
- the learning data L n used here is an m ⁇ (k + 1) matrix in which B n (0) and the data of columns 1 to k extracted from the storage data C n are combined, and is expressed by the following equation. Represented by.
- the data in column 0 (first column) is B n (0) this time
- the data in column k (k + 1st column) is B n (k) k times before.
- k is determined by trial and error so as to satisfy the learning time for updating the learned information at an arbitrary timing and to secure the desired monitoring accuracy.
- the abnormality degree calculation unit 23 scales (normalizes) the time window data B n based on the learned information, and calculates the scaled data B s, n (step S4 in FIG. 2).
- Scaled data B s, n is the following formula, is calculated by calculating all the matrix elements of B S.
- min n is the minimum value of the data in the nth signal in the learned information.
- max n is the maximum value of the data in the nth signal in the learned information.
- B s and n are data (m ⁇ 1 matrix) obtained by scaling B n.
- [B n ] i is the i-th scalar quantity of B n.
- [B s, n ] i is the i-th scalar quantity of B s, n. All the vector elements of the scaled data Bs , n fall within the range of the minimum value 0 to the maximum value 1.
- the abnormality degree calculation unit 23 can handle a plurality of vehicle information having different orders in the same manner and calculate the abnormality degree by such a scaling process.
- the anomaly degree calculation unit 23 reconstructs the scaled data Bs and n by self-coding using the self-encoder 23a (see FIG. 7) (step S5 in FIG. 2). ..
- the self-encoder 23a includes an input layer 23a1 to which input data is input, an output layer 23a2 to which reconstruction data is output, and an intermediate layer (hidden layer) 23a3 provided between the input layer 23a1 and the output layer 23a2. , Equipped with.
- the latent variable z and the reconstruction data x ⁇ in the intermediate layer 23a3 are calculated by the following equations, respectively.
- x is a matrix input to the self-encoder 23a.
- W x (1) , b x (1) , W x (2) , and b x (2) are self-encoder information corresponding to the input x.
- z is a latent variable
- x ⁇ ( ⁇ on x) is a matrix (reconstruction data) output from the self-encoder 23a.
- the abnormality degree calculation unit 23 calculates the error (reconstruction error) e included in the reconstruction data (step S6 in FIG. 2).
- the reconstruction error e is calculated by the following formula.
- ⁇ x is a difference
- e is the Euclidean norm of ⁇ x.
- the abnormality degree calculation unit 23 can collectively process them to calculate the reconstruction error.
- the numbers of the vehicle signals having a correlation are a and b
- the input data and x and the reconstruction data x ⁇ are expressed by the following equations.
- the anomaly degree calculation unit 23 calculates the reconstruction error e a from B s, a and B ⁇ s, a ( ⁇ on B), and B s, b and B ⁇ s, b (on B). ⁇ ) to calculate the reconstruction error e b from.
- the self-encoder information is Wab (1) , bab (1) , Wab (2) , bab (2) .
- the self-encoder 23a is not limited to one that uses the sigmoid function ⁇ as the activation function, and may use, for example, a hyperbolic tangent function or the like. Further, the intermediate layer 23a in the self-encoder is not limited to one layer, and a plurality of layers may be laminated. Further, the method for calculating the reconstruction error is not limited to the one using the Euclidean norm (L2 norm), and may be, for example, the one using the L1 norm or the like.
- L2 norm Euclidean norm
- the reconstruction method is not limited to the one using the self-encoder 23a, and may be, for example, the one using principal component analysis or the like. Since the vehicle information at the time of determination is almost only normal data, the abnormality degree calculation unit 23 cannot perform processing using another method using a neural network such as classification learning. On the other hand, the abnormality degree calculation unit 23 extracts latent variables using the self-encoder 23a that has learned only normal data, and reconstructs the data based on the extracted latent variables. The anomaly degree calculation unit 23 generates normal waveform data learned by reconstruction even when there is anomalous data deviating from the waveform data. Therefore, the abnormality degree calculation unit 23 calculates the reconstruction error e as information deviating from the normal waveform data.
- error probability calculation unit 23 a learned information determination function f n (x), and, on the basis of the reconstruction error e n, scores of the n-th vehicle information (error probability) is calculated S n (Step S7 in FIG. 2), the calculation result is output to the abnormality determination unit 24.
- the degree of abnormality Sn is calculated by the following formula.
- the anomaly degree calculation unit 23 uses a one-class SVM (Support Vector Machine) as the determination function f n (x). Therefore, the score S n, determined function f n (x) takes a value greater than or equal to zero if the normal value is entered, the decision function f n minus if the abnormal value is input to the (x) Take a value.
- SVM Serial Vector Machine
- Abnormality determination unit 24 the abnormality score (error probability) calculated by the calculating unit 23 obtains the S n, based on the obtained scores, determines whether the vehicle signal is abnormal (in FIG. 2 Step S8).
- the abnormality determination unit 24 determines that the corresponding vehicle signal is abnormal when a state in which the score Sn is negative occurs continuously for a plurality of time windows, and responds in other cases. It is determined that the vehicle signal is normal.
- the abnormality determination unit 24 outputs the determination result to the notification device 30 (step S9 in FIG. 2).
- the abnormality determination unit 24 can prevent the vehicle signal from being determined as an abnormality when pulse noise is generated, and can detect a true abnormality.
- the learning unit 25 learns in real time in parallel with the operations of the sampling rate adjustment unit 21, the time window processing unit 22, the abnormality degree calculation unit 23, and the abnormality determination unit 24. And generate learned information.
- the learning unit 25 extracts the maximum value max n and the minimum value min n included in the learning data L n (step S12 in FIG. 3). Subsequently, the learning unit 25 scales the learning data L n based on the extracted maximum value max n and the minimum value min n , and calculates the scaled learning data L s, n. Subsequently, the learning unit 25 learns the self-encoder 23a using the learned data Ls and n , and the self-encoder information W x (1) , b x (1) , W x (2). , B x (2) is generated (step S12 in FIG. 3).
- a scaling conjugate gradient descent method can be mentioned.
- a preset value is used as the hyperparameter.
- the learning unit 25 calculates the reconstruction error based on the self-encoder information W x (1) , b x (1) , W x (2) , b x (2) , and recalculates the calculated error.
- the decision function f n (x) is generated with the configuration error as normal data (step S12 in FIG. 3).
- the learning unit 25 has a maximum value max n and a minimum value min n , self-encoder information W x (1) , b x (1) , W x (2) , b x (2) , and a determination function f n (x). ) Is stored in the storage 20b as learned information.
- the learning unit 25 learns independently of the processing of the sampling rate adjustment unit 21, the time window processing unit 22, the abnormality degree calculation unit 23, and the abnormality determination unit 24, it is separate from the abnormality detection device 20 ( It can also be embodied by another processor, computer (PC), etc.).
- the update unit 26 reads the updated learned information and stores it in the volatile memory 20a (step S13 in FIG. 3).
- the notification device 30 is composed of a display capable of outputting an image, a speaker capable of outputting voice, and the like, and acquires a determination result by the abnormality determination unit 24 and notifies the user.
- the abnormality detection device 20 has a time window processing unit 22 for extracting data included in the time window by dividing a plurality of vehicle signals including time series data by a time window, and the time window.
- the learning unit 25 that generates learned information by learning using the time-series data of the vehicle signal separated by, and the time-series data of the vehicle signal separated by the time window are based on the learned information. It is characterized by including an abnormality degree calculation unit 23 that reconstructs, calculates an error before and after the reconstruction, and calculates an abnormality degree based on the calculated error and the learned information.
- the abnormality detection device 20 can calculate the degree of abnormality of the vehicle information delimited by the time window by using the learned information using the vehicle information delimited by the time window before that. That is, the abnormality detection device 20 does not need to prepare the learned information in advance of the detection work, and can suitably detect the abnormality based on the learned information updated in real time.
- the abnormality degree calculation unit 23 uses the time series data of the vehicle signal separated by the time window this time and the time series of the vehicle signal separated by the time window before the previous time. It is characterized in that the degree of abnormality is calculated based on the learned information using the data. Therefore, the abnormality detection device 20 can more preferably execute the generation of learned data and the calculation of the degree of abnormality in parallel.
- the learned information includes the maximum value and the minimum value of the time series data of the vehicle signal separated by the time window before the previous time
- the abnormality degree calculation unit includes the maximum value.
- the time-series data of the vehicle signal separated by the time window of this time is scaled, and the scaled time-series data of the vehicle signal is self-encoded included in the learned information. It is characterized by being reconstructed by a vessel 23a. Therefore, the abnormality detection device 20 can handle a plurality of vehicle information having different orders in the same manner and appropriately calculate the degree of abnormality.
- the abnormality detection device 20 includes a sampling rate adjusting unit 21 for matching the sampling rates of the plurality of vehicle signals, and the time window processing unit 22 has time-series data of the plurality of vehicle signals whose sampling rates have been adjusted. Is characterized by being separated by the time window. Therefore, the abnormality detection device 20 can improve the correlation of each vehicle signal by matching the sampling rates of the plurality of vehicle information, and can prevent over-learning of the vehicle signal having a large sampling rate. ..
- the abnormality detection device 20 has an abnormality determination unit 24 that determines that the vehicle signal is abnormal when an abnormality occurs continuously in a predetermined number of the time windows based on the calculated abnormality degree. It is characterized by being prepared. Therefore, the abnormality detection device 20 can prevent the vehicle signal from being determined as an abnormality when pulse noise is generated, and can detect a true abnormality.
- the present invention is not limited to the above-described embodiments and can be appropriately modified without departing from the gist of the present invention.
- the abnormality detection device 20 may be configured to omit the abnormality determination unit 24.
- the calculation result of the abnormality degree calculation unit 23 is output to the notification device 30 by voice or image.
- the present invention can also be realized as an abnormality detection program that causes the computer to function as the abnormality detection device 20.
- Anomaly detection system 10 Measuring device 20 Anomaly detection device 21 Sampling rate adjustment unit 22 Time window processing unit 23 Abnormality calculation unit 24 Abnormality determination unit 25 Learning unit
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Automation & Control Theory (AREA)
- General Physics & Mathematics (AREA)
- Software Systems (AREA)
- Mathematical Physics (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Computing Systems (AREA)
- Mechanical Engineering (AREA)
- Transportation (AREA)
- Human Computer Interaction (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Medical Informatics (AREA)
- Testing And Monitoring For Control Systems (AREA)
- Traffic Control Systems (AREA)
- Testing Of Devices, Machine Parts, Or Other Structures Thereof (AREA)
Abstract
Description
また、本発明の異常検知プログラムは、コンピュータを前記異常検知装置として機能させることを特徴とする。
計測装置10は、台上耐久試験における模擬的な試験走行中の車両2の各種信号(車両信号)を計測し、計測された複数の車両信号を異常検知装置20へ送信する。計測される車両信号は、オーダ及び/又は単位が異なる信号を含み、車輪の回転速度(車輪速)、エンジンの回転速度、エンジンの回転トルク、エンジンのスロットル開度、エンジンの各部温度(排気温、水温、油温)、駆動モータの出力トルク、駆動モータの発生電流、エンジン制御ユニットからの故障コード等が例として挙げられる。
異常検知装置20は、計測装置10によって計測された複数の車両信号を取得し、取得された車両信号ごとに異常を検知する。異常検知装置20は、例えば、CPU(Central Processing Unit)、ROM(Read-Only Memory)、RAM(Random Access Memory)、入出力回路等によって構成されている。異常検知装置20は、機能部として、サンプリングレート調整部21と、時間窓処理部22と、異常度算出部23と、異常判定部24と、学習部25と、を備える。
サンプリングレート調整部21は、計測装置10によって計測された複数(n種類)の車両信号を取得し、取得された複数の車両信号のサンプリングレートを調整して一致させる(図2のステップS2)。図4に示すように、複数の車両信号は、それぞれ異なるサンプリングレートごとの時系列データを含む。図4において、各車両信号に含まれる信号は、時刻を示す軸上の黒丸によって示されている。サンプリングレート調整部21は、予め記憶された所定サンプリングレートに基づいて、車両信号内のデータのうち、所定サンプリングレートのタイミング(時間間隔tごと)のそれぞれに最も近いデータ(図4において、点線丸印で囲まれたデータ)を、当該タイミングにおけるデータとして抽出するとともに、それ以外のデータを間引く。サンプリングレート調整部21は、サンプリングレートが調整された車両信号を時間窓処理部22へ出力する。ここで、所定サンプリングレートは、複数の車両信号のサンプリングレートのうち、最も大きいもの以上であることが望ましい。
時間窓処理部22は、揮発性メモリ20a上に実装されている。時間窓処理部22は、サンプリングレート調整部21によってサンプリングレートが調整されたn種類の車両信号を取得し、サンプリングレートが調整された車両信号から、時間窓に含まれる所定個数(m個)のデータBnを抽出する(図2のステップS3)。ここで、tは、時間窓のスライド幅であるとともに、異常判定の周期(ステップS2~S9の実行に要する時間)である。また、Bn(又はBn (0))は、今回の時間窓に含まれるデータ(m×1行列)である。
図1に示すように、異常度算出部23は、揮発性メモリ20aに記憶された学習済み情報を取得する(図2のステップS1)とともに、時間窓処理部22によって生成されたBnを取得し、取得されたBn及び学習済み情報に基づいて、車両信号の異常度を算出する。学習済み情報に関しては、後で詳細に説明する。
異常判定部24は、異常度算出部23によって算出されたスコア(異常度)Snを取得し、取得されたスコアに基づいて、車両信号が異常であるか否かを判定する(図2のステップS8)。異常判定部24は、スコアSnがマイナスである状態が複数の時間窓に対して連続して発生した場合に、対応する車両信号は異常であると判定し、それ以外の場合には、対応する車両信号は正常であると判定する。異常判定部24は、異常であると判定した場合に、判定結果を通知装置30へ出力する(図2のステップS9)。
学習部25は、ストレージ20aに記憶された保存データに基づいて、サンプリングレート調整部21、時間窓処理部22、異常度算出部23及び異常判定部24の動作と並行して、リアルタイムで学習を行い、学習済み情報を生成する。学習部25は、ストレージ20bに記憶されたCnのうち、学習に必要な時間の行列を読み出し、読み出されたデータを学習用データLnとする(図3のステップS11)。
通知装置30は、画像出力可能なディスプレイ、音声出力可能なスピーカ等によって構成されており、異常判定部24による判定結果を取得してユーザに通知する。
したがって、異常検知装置20は、時間窓で区切られた車両情報の異常度を、それよりも前の時間窓で区切られた車両情報を用いた学習済み情報を用いて算出することができる。すなわち、異常検知装置20は、学習済み情報を検知作業よりも事前に準備する必要が無く、リアルタイムで更新された学習済み情報に基づいて、異常を好適に検知することができる。
したがって、異常検知装置20は、学習済みデータの生成及び異常度の算出をより好適に並行して実行することができる。
したがって、異常検知装置20は、オーダが異なる複数の車両情報を同様に取り扱って異常度を好適に算出することができる。
したがって、異常検知装置20は、複数の車両情報のサンプリングレートを一致させることによって、各車両信号の相関性を向上することができるとともに、サンプリングレートが大きい車両信号の過学習を防止することができる。
したがって、異常検知装置20は、車両信号にパルス的ノイズが発生した場合に異常と判定されることを防止し、真の異常を検知することができる。
10 計測装置
20 異常検知装置
21 サンプリングレート調整部
22 時間窓処理部
23 異常度算出部
24 異常判定部
25 学習部
Claims (6)
- 時系列データを含む複数の車両信号を時間窓で区切ることによって、当該時間窓に含まれるデータを抽出する時間窓処理部と、
前記時間窓で区切られた前記車両信号の時系列データを用いた学習によって、学習済み情報を生成する学習部と、
前記時間窓で区切られた前記車両信号の時系列データを前記学習済み情報に基づいて再構成し、再構成前後の誤差を算出するとともに、算出された前記誤差と前記学習済み情報とに基づいて異常度を算出する異常度算出部と、
を備えることを特徴とする異常検知装置。 - 前記異常度算出部は、今回の前記時間窓によって区切られた前記車両信号の時系列データと、前回以前の前記時間窓によって区切られた前記車両信号の時系列データを用いた前記学習済み情報と、に基づいて、前記異常度を算出する
ことを特徴とする請求項1に記載の異常検知装置。 - 前記学習済み情報は、前回以前の前記時間窓によって区切られた前記車両信号の時系列データの最大値及び最小値を含み、
前記異常度算出部は、前記最大値及び前記最小値に基づいて、今回の前記時間窓によって区切られた前記車両信号の時系列データをスケーリングし、スケーリングされた前記車両信号の時系列データを、前記学習済み情報に含まれる自己符号化器によって再構成する
ことを特徴とする請求項2に記載の異常検知装置。 - 複数の前記車両信号のサンプリングレートを一致させるサンプリングレート調整部を備え、
前記時間窓処理部は、サンプリングレートが調整された複数の前記車両信号の時系列データを、前記時間窓で区切る
ことを特徴とする請求項1から請求項3のいずれか一項に記載の異常検知装置。 - 算出された前記異常度に基づいて、所定数の前記時間窓に連続して異常が発生した場合に、前記車両信号は異常であると判定する異常判定部を備える
ことを特徴とする請求項1から請求項4のいずれか一項に記載の異常検知装置。 - コンピュータを請求項1から請求項5のいずれか一項に記載の異常検知装置として機能させる異常検知プログラム。
Priority Applications (3)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| MX2023000909A MX2023000909A (es) | 2020-07-20 | 2021-06-18 | Dispositivo de deteccion de anomalias y programa de deteccion de anomalias. |
| CN202180049128.5A CN115989465B (zh) | 2020-07-20 | 2021-06-18 | 异常检测装置及异常检测程序 |
| US18/005,897 US12311962B2 (en) | 2020-07-20 | 2021-06-18 | Abnormality detection device and abnormality detection program |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP2020124062A JP7072611B2 (ja) | 2020-07-20 | 2020-07-20 | 異常検知装置及び異常検知プログラム |
| JP2020-124062 | 2020-07-20 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2022019020A1 true WO2022019020A1 (ja) | 2022-01-27 |
Family
ID=79729448
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/JP2021/023245 Ceased WO2022019020A1 (ja) | 2020-07-20 | 2021-06-18 | 異常検知装置及び異常検知プログラム |
Country Status (5)
| Country | Link |
|---|---|
| US (1) | US12311962B2 (ja) |
| JP (1) | JP7072611B2 (ja) |
| CN (1) | CN115989465B (ja) |
| MX (1) | MX2023000909A (ja) |
| WO (1) | WO2022019020A1 (ja) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115656847A (zh) * | 2022-12-08 | 2023-01-31 | 东莞先知大数据有限公司 | 一种电动车电池异常确定方法、电子设备和存储介质 |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3803529A4 (en) * | 2018-06-10 | 2022-01-19 | OSR Enterprises AG | SYSTEM AND METHOD FOR IMPROVING SENSOR OPERATION IN A VEHICLE |
| JP7310754B2 (ja) * | 2020-08-21 | 2023-07-19 | いすゞ自動車株式会社 | 診断装置 |
| KR20250070415A (ko) * | 2023-11-13 | 2025-05-20 | 에이치엘로보틱스 주식회사 | 차량 상태 모니터링 장치 및 방법 |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2014234113A (ja) * | 2013-06-04 | 2014-12-15 | 株式会社デンソー | 車両用基準値生成装置 |
| JP2019175462A (ja) * | 2018-03-29 | 2019-10-10 | コリア インスティテュート オブ オーシャン サイエンス テクノロジー | 教師なし学習方法を用いた船舶の異常運航状態自動識別システムおよびその方法 |
Family Cites Families (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP6361175B2 (ja) | 2014-03-06 | 2018-07-25 | 株式会社豊田中央研究所 | 異常診断装置及びプログラム |
| JP6402541B2 (ja) | 2014-08-26 | 2018-10-10 | 株式会社豊田中央研究所 | 異常診断装置及びプログラム |
| JP6839963B2 (ja) * | 2016-01-08 | 2021-03-10 | パナソニック インテレクチュアル プロパティ コーポレーション オブ アメリカPanasonic Intellectual Property Corporation of America | 異常検知方法、異常検知装置及び異常検知システム |
| JP6969993B2 (ja) | 2017-12-01 | 2021-11-24 | 株式会社デンソー | 情報抽出装置 |
| CN109101986A (zh) | 2018-06-07 | 2018-12-28 | 国网山东省电力公司青岛供电公司 | 基于栈式降噪自编码器的输变电设备状态异常检测方法和系统 |
| CN110034968A (zh) * | 2019-03-12 | 2019-07-19 | 上海交通大学 | 基于边缘计算的多传感器融合车辆安全异常检测方法 |
-
2020
- 2020-07-20 JP JP2020124062A patent/JP7072611B2/ja active Active
-
2021
- 2021-06-18 MX MX2023000909A patent/MX2023000909A/es unknown
- 2021-06-18 WO PCT/JP2021/023245 patent/WO2022019020A1/ja not_active Ceased
- 2021-06-18 CN CN202180049128.5A patent/CN115989465B/zh active Active
- 2021-06-18 US US18/005,897 patent/US12311962B2/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2014234113A (ja) * | 2013-06-04 | 2014-12-15 | 株式会社デンソー | 車両用基準値生成装置 |
| JP2019175462A (ja) * | 2018-03-29 | 2019-10-10 | コリア インスティテュート オブ オーシャン サイエンス テクノロジー | 教師なし学習方法を用いた船舶の異常運航状態自動識別システムおよびその方法 |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115656847A (zh) * | 2022-12-08 | 2023-01-31 | 东莞先知大数据有限公司 | 一种电动车电池异常确定方法、电子设备和存储介质 |
Also Published As
| Publication number | Publication date |
|---|---|
| JP7072611B2 (ja) | 2022-05-20 |
| CN115989465B (zh) | 2025-11-14 |
| MX2023000909A (es) | 2023-02-22 |
| JP2022020512A (ja) | 2022-02-01 |
| US12311962B2 (en) | 2025-05-27 |
| US20230322240A1 (en) | 2023-10-12 |
| CN115989465A (zh) | 2023-04-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2022019020A1 (ja) | 異常検知装置及び異常検知プログラム | |
| US12607995B2 (en) | Automated analysis of non-stationary machine performance | |
| Lei et al. | Fault diagnosis of wind turbine based on Long Short-term memory networks | |
| JP6740247B2 (ja) | 異常検出システム、異常検出方法、異常検出プログラム及び学習済モデル生成方法 | |
| CN110276416A (zh) | 一种滚动轴承故障预测方法 | |
| JP2021528745A (ja) | アプリケーション情報に関連する時系列ズデータに関する深層学習を使用した異常検出 | |
| WO2022207223A1 (en) | Method and system for optimizing a digital twin simulation model using machine learning | |
| US20190332936A1 (en) | Additional learning method for deterioration diagnosis system | |
| JP7238378B2 (ja) | 異常検出装置、異常検出プログラム、及び、異常検出方法 | |
| KR102546340B1 (ko) | 노이즈 필터를 이용한 이상치 탐지 방법 및 장치 | |
| EP2771846A1 (en) | Heterogeneous data fusion using gaussian processes | |
| CN116720073A (zh) | 一种基于分类器的异常检测提取方法与系统 | |
| Choudhury | Prognosticating autism spectrum disorder using artificial neural network: Levenberg-Marquardt algorithm | |
| CN111079348B (zh) | 一种缓变信号检测方法和装置 | |
| US12099352B2 (en) | Methods and systems for fault diagnosis | |
| Yu et al. | A novel 1D‐CNN‐based diagnosis method for a rolling bearing with dual‐sensor vibration data fusion | |
| JP2020149259A (ja) | 異常診断装置 | |
| Anacur et al. | Handwriting analysis for Alzheimer’s diagnosis: feature enrichment and machine learning classification | |
| JP7281394B2 (ja) | 異常診断装置およびプログラム | |
| CN114881110A (zh) | 一种在轨航天器舱内总压变化模式的实时检测方法 | |
| Pandey et al. | Tool wear estimation using few-shot learning in deep neural network architectures | |
| Patil et al. | Algorithms for Detection of Anomalies in Data From Clinical Studies and Medical Registries | |
| Munim et al. | Prediction accuracy in maritime simulator training performance assessment with varying data frequency | |
| CN120216968B (zh) | 一种基于前端感知数据的预警方法及系统 | |
| Zhang et al. | Bearing performance degradation assessment based on a combination of multi-scale entropy and K-medoids clustering |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 21846218 Country of ref document: EP Kind code of ref document: A1 |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 202180049128.5 Country of ref document: CN |
|
| NENP | Non-entry into the national phase |
Ref country code: DE |
|
| 122 | Ep: pct application non-entry in european phase |
Ref document number: 21846218 Country of ref document: EP Kind code of ref document: A1 |
|
| WWG | Wipo information: grant in national office |
Ref document number: 18005897 Country of ref document: US |
|
| WWG | Wipo information: grant in national office |
Ref document number: MX/A/2023/000909 Country of ref document: MX |

