CN117601926A

CN117601926A - Cooperative monitoring device and system of traction electric transmission and safety monitoring equipment

Info

Publication number: CN117601926A
Application number: CN202311586041.9A
Authority: CN
Inventors: 张和生; 刘恒志; 汤昳琮; 姚豫强; 丁卓; 要思敏
Original assignee: Beijing Jiaotong University
Current assignee: Beijing Jiaotong University
Priority date: 2023-11-24
Filing date: 2023-11-24
Publication date: 2024-02-27

Abstract

The invention provides a cooperative monitoring device and a cooperative monitoring system for rail train traction electric transmission and safety monitoring equipment by utilizing edge calculation. The edge calculation device includes: the device comprises a sensor module, a positioning module, an edge master control module, a storage module and various communication interface modules; the sensor module preprocesses the sensor signals to obtain normal state quantity and/or abnormal quantity, and sends the normal state quantity and/or abnormal quantity to the edge master control module through the internal signal transmission bus; the edge master control module comprises a DCPU, an MCPU, a data storage array, a Flash memory and a bus interface; the DCPU extracts characteristics of the normal state, analyzes and judges the abnormal state and stores the result; the MCPU realizes the initialization of the edge computing device, various bus transmission and the management of various modules. The device can exchange data with other edge computing devices through Ethernet to form a collaborative monitoring system, realize collaborative monitoring of traction electric transmission and safety detection equipment of the rail train, and conduct fault prediction and health management on the traction electric transmission and safety detection equipment.

Description

Cooperative monitoring device and system of traction electric transmission and safety monitoring equipment

Technical Field

The invention relates to the technical field of rail train monitoring, in particular to a collaborative monitoring device and a collaborative monitoring system for traction electric transmission systems and safety monitoring system equipment by utilizing edge calculation.

Background

Rail trains are one of the important tasks as mass public transportation vehicles, the safe operation of which. Traction drive system relates to a plurality of fields such as power electronics, motor, microcomputer control, mainly includes: traction converter and control device TCU (Traction Control Unit ), network communication and control device, monitoring device, traction motor, etc., traction transmission and its control system embody the technical advanced level of rail train equipment to a certain extent.

Traction electric drive control typically employs a network control system that integrates train monitoring, diagnostics, and control, collectively referred to as a TCN (Train Communication Network ), for control, detection, and diagnostics of rail train traction drive equipment. The TCN includes a WTB (wireless Train Bus) controlled by a Train level network and an MVB (Multi-function Vehicle Bus, multifunctional communication Bus) controlled by a vehicle level network. The WTB is used for connecting the vehicle equipment which can be dynamically grouped in each section, and the transmission of process data and message data is realized. The MVB is used for connecting fixed equipment in the vehicle to realize the transmission of control instructions and state data. The state monitoring of the traction electric transmission system is also based on TCN, and has important significance for ensuring the safe operation of the train.

The monitoring system of the rail train in the prior art comprises two main types of state monitoring facing traction transmission control and state monitoring facing operation safety. The state detection for traction transmission control comprises secondary side voltage of a transformer, motor current, intermediate direct current voltage, motor rotating speed and the like, and the detection systems are connected with the TCU through MVB or directly. The state monitoring of traction drive control mainly comprises voltage and current monitoring, including monitoring of input voltage, direct current voltage, stator three-phase voltage, output current, rotating speed signals and the like of a traction converter, estimating motor magnetic flux and torque, and speed-free direct torque control can be free of speed signals. The vector control system monitors three-phase voltages, three-phase currents, and motor rotational speeds for vector control. Both the monitoring sensor and the process operate within the traction converter.

The safety-oriented state monitoring comprises a smoke and fire alarm system, a bearing temperature monitoring system, a instability monitoring system, a speed monitoring system and the like. The bearing temperature monitoring system is characterized in that temperature sensors are arranged on a bogie bearing, an axle box bearing and a traction motor bearing and are used for detecting the temperature of the bearing in real time when a train runs, and automatically alarming or taking a stopping measure when the temperature exceeds the limit. The instability monitoring system is to deploy an acceleration sensor for detecting vibration on the bogie, monitor the transverse vibration acceleration of the bogie and judge whether the instability is possible or not; and the sensor data are transmitted to the control unit, the acceleration is analyzed, and when the bogie is judged to have the signs of instability, an alarm is sent out, and a deceleration measure is adopted. The speed monitoring system is characterized in that a speed sensor is arranged on a front end cover of the axle circumference, and the speed sensor is used for providing a speed signal for a braking system and providing a speed signal for an ATP (Automatic Train Protection, automatic train protection system) to control the speed of a train.

The current TCN data transmission rate is relatively low, and the transmission of control instructions and a small amount of state data is realized during design. The need for broadband data transmission by intelligent train and unmanned train technological development is not considered; the current monitoring system mainly discovers abnormal values, the data transmission quantity of the sensors is large, the current train communication network cannot bear more data transmission quantity, the increase of the deployment quantity of the sensors is limited, and the quantity and the types of the accessed sensors are small. New rail trains have employed TCN + ethernet for traction electric drive system control and ethernet for monitoring the status of the equipment.

Currently, the drawbacks of the collaborative monitoring method between the electric transmission equipment and the safety monitoring equipment of the rail train in the prior art include:

1) The collaborative monitoring cannot be realized: the existing monitoring systems of the rail trains are relatively independent, different monitoring quantities cannot be interacted, and the problem that equipment runs safely cannot be discovered through cooperative monitoring is solved.

The existing independent monitoring mode cannot discover degradation process and faults of traction electric drive system equipment states in time, and cannot cooperatively monitor the equipment to arrange operation. If the motor current reaches the rated current or exceeds the rated current, the shaft temperature can rapidly fail when the shaft temperature approaches the failure threshold value, and the current method cannot monitor the current. The existing monitoring mode cannot form a collaborative monitoring system, so that the omission ratio is high.

2) Lack of edge computing devices suitable for collaborative monitoring: the monitoring of the existing rail train traction electric transmission system aims at finding abnormality, so that the host computer has limited computing capacity, weak data processing capacity and few network interfaces; the vehicle-mounted board has small storage capacity and small number of accessed sensors, a large amount of sensor data needs to be transmitted through a vehicle-mounted network, and the number and types of the accessed sensors are small due to the limited bandwidth of the vehicle-mounted network. The existing monitoring system vehicle-mounted board cards cannot interact with each other and cannot realize collaborative monitoring.

3) Lack of intelligent online analysis methods: the existing monitoring mode judges abnormality through a fixed threshold value, cannot record the state of electric transmission equipment and an evolution rule thereof, cannot realize a complex decision process at the equipment side in the way, and causes the problems of high false alarm rate and high omission rate.

Smart trains and unmanned trains are in urgent need of failure prediction and health management (PHM, prognostics Health Management) systems, requiring the addition of deployment sensors on some devices, and the addition of sensor types that have not been deployed before to monitor the status of the devices. At present, data monitored by a sensor is directly transmitted to a vehicle-mounted host through a vehicle-mounted network. With the increase of sensor monitoring nodes, the transmission data volume is greatly increased, and the real-time performance and the network capacity can not meet the demands. The current train communication network cannot bear more data transmission quantity, and the increase of the sensor deployment quantity is limited.

The PHM system needs to master the normal running state data and the characteristics thereof and the law of the evolution of the equipment state under different running environment conditions. Existing monitoring systems lack these functionality and are difficult to apply to state-based equipment maintenance and repair.

Current state monitoring of rail train electric drive systems is aimed at discovering anomalies. A fixed threshold is adopted as a standard of abnormality detection, for example, a shaft temperature monitoring system only measures temperature and judges with the fixed threshold; the voltage and current system is used for train control and the like, and alarms when the set threshold value is exceeded. This results in a high false positive rate.

Disclosure of Invention

The embodiment of the invention provides a collaborative monitoring device and a collaborative monitoring system for a traction electric drive system by utilizing edge calculation, so as to realize effective fault monitoring for the traction electric drive system.

In order to achieve the above purpose, the present invention adopts the following technical scheme.

According to an aspect of the present invention, there is provided an edge computing device including: the device comprises a power module, a sensor module, an edge master control module and a storage module which are mutually connected through an internal signal transmission bus;

the sensor module comprises a sensor, conditioning conversion, processing judgment, storage and communication, wherein different conditioning and conversion circuits are adopted according to different output signals of the sensor, the sensor signals are preprocessed and the threshold value is judged to obtain normal state quantity and/or abnormal quantity of a monitored object, and the normal state quantity and/or abnormal quantity of the monitored object is sent to the edge master control module through an internal signal transmission bus interface; the edge master control module comprises a data processing central processing unit (DCPU), an edge device Management Central Processing Unit (MCPU), a data storage array, a Flash memory and a bus interface, wherein the DCPU gathers data of one or more sensor modules, acquires characteristic quantity of a normal state and state variation quantity reflecting evolution of the normal state by utilizing a characteristic extraction algorithm for the normal state quantity, learns and models normal data or normal data characteristics, and obtains a variation rule of the normal state; analyzing and further judging the abnormal quantity, combining the normal data characteristics and the evolving state change rate, performing fault diagnosis on the abnormal quantity, and determining a threshold revision quantity for frequent faults according to the fault diagnosis result; the MCPU realizes the initialization of the edge computing device, the internal bus transmission, the management of various communication modules and various modules, and the cooperative address allocation of different edge computing devices when the train is in initial operation; the DCPU and the MCPU exchange related data through a Flash memory, and control and data exchange are carried out between the DCPU and other modules of the edge computing device through an internal bus interface;

The storage module is used for configuring a storage chip array, managing a CPU and a bus interface, storing normal operation characteristics and evolution characteristics thereof, key data points, abnormal data, fault diagnosis models and parameters, fault diagnosis results, threshold values and data characteristics and key data points of multi-source collaborative monitoring, and carrying out data interaction with the edge master control module through an internal data transmission bus;

the power module is used for taking power from a vehicle-mounted power supply, obtaining a direct-current power supply through DC-DC conversion and supplying power to other modules by using a power bus.

Preferably, the device further comprises: the system comprises a positioning module, a multifunctional communication bus MVB interface module, an Ethernet interface module and a wireless interface module;

the MVB interface module is used as a main interface of the train communication network TCN accessed by the edge computing device, receives a control instruction of the central control unit CCU, and reports the key abnormal point to the CCU through the TCN;

the Ethernet interface module is used as a communication network for connecting the edge computing device with the train monitoring center TMC, so as to realize the mutual interaction between the edge computing devices;

the wireless interface module is used for controlling the access, networking and address allocation of the vehicle-mounted wireless sensor, carrying out data transmission with the ground monitoring data center and sending the processing result of the vehicle-mounted edge computing device to the ground monitoring center;

The positioning module is used for providing position data for the edge computing device by adopting a satellite positioning chip and providing position data for the edge master control module through an internal signal transmission bus.

According to one aspect of the present invention, there is provided a collaborative monitoring system of traction electric drive and safety monitoring equipment based on the edge computing device, comprising: an edge computing device and a sensor module are arranged in each carriage of the rail train and are used for carrying out multi-source physical quantity collaborative monitoring on traction electric transmission equipment and safety monitoring equipment, the edge computing devices of each rail train are connected with each other through Ethernet and are connected to a train monitoring center TMC, the TMC receives related data, carries out model training and parameter determination, determines various thresholds, carries out fault prediction and health management and issues results and tasks;

the edge computing device is accessed into the TCN through the vehicle gateway by utilizing the MVB interface, is connected with the CCU, reports abnormality to the CCU and receives a control instruction sent by the CCU.

Preferably, the collaborative monitoring system comprises a car layer, an edge layer and an end layer;

the end layer comprises a sensor and a sensor module, the sensor is arranged at the traction electric transmission equipment and/or the safety monitoring equipment of the train, the type and the number of the sensors are selected according to the characteristics of the physical quantity of the monitored equipment, the state of the monitoring equipment is acquired, the sensor is connected with the sensor module, the sensor module preprocesses the data, the threshold value is judged and early warned, and the judging result is sent to the edge computing device; receiving an instruction sent by an edge computing device; the sensor module is deployed at the monitoring equipment together with the sensor or the sensor is deployed at the monitoring equipment according to the monitoring task and the deployment site, and the sensor module is deployed in the edge computing device;

The boundary layer comprises an edge computing device which is arranged on a train, receives temperature, vibration and speed data of a traction motor, a bearing which are arranged on a bogie respectively, performs data aggregation on data transmitted by one or more sensor modules, acquires characteristic quantity of a normal state and state change quantity reflecting the evolution of the normal state by utilizing a characteristic extraction algorithm on the normal state quantity, learns and models the normal data or the normal data characteristic, obtains the change rule of the normal state, and stores the change rule; analyzing and further judging abnormal quantity combined with normal data characteristics and the evolving state change rate thereof, performing fault diagnosis on the abnormal quantity, determining a threshold value revision quantity of a frequently-occurring fault according to a fault diagnosis result, storing the result, and receiving control instructions sent by TMC of a vehicle layer by different edge computing devices through interactive data of an Ethernet network;

the vehicle layer comprises TMC, and is connected with edge computing devices of all carriages through Ethernet to form a vehicle-mounted collaborative monitoring system, and performs data interaction with a ground monitoring center through a general packet radio service GPRS or a mobile communication network to perform model training and parameter determination of the same physical quantity, and receives monitoring and analysis results of the same physical quantity from different edge computing devices and compares and judges the results; model training and parameter determination of different physical quantity collaborative monitoring are carried out, monitoring and analysis results of different physical quantities monitored in a collaborative state are received, and comparison and discrimination are carried out; determining an abnormality discrimination threshold and a dynamic threshold of different rail physical quantities, and judging the state of equipment according to an algorithm related to equipment operation fault prediction and health management according to the result of an edge computing device of a vehicle; and receiving an instruction sent by the ground monitoring center.

Preferably, the collaborative monitoring system is used for performing normal shaft temperature data feature learning by adopting a long-short-term memory network LSTM model through a vehicle layer TMC, and establishing a mapping relationship between shaft temperature historical data and current data through a deep learning algorithm, and the specific processing process comprises the following steps:

LSTM introduces a new internal state c in each cell _t Linear circulation information transmission is carried out, and information is output to the external state h of the hidden layer _t By introducing forgetting door f _t Input gate i _t And an output gate o _t Controlling the information quantity;

the vehicle layer TMC performs standardization and sliding window processing on normal shaft temperature data, the preprocessed shaft temperature sequence is input into an LSTM model for training, a data set required by model training is from historical normal shaft temperature data stored by the TMC or normal shaft temperature data transmitted by a boundary layer edge computing device, and the mapping relation between the shaft temperature historical data and current data is obtained, namely:

temp _t ＝f(temp _t-1 ,temp _t-2 ,…,temp _t-m ) (1)

in the formula, temp _t The shaft temperature at time t; f (·) represents a nonlinear mapping relationship, which refers to a black box model without an explicit expression; m is the length of the model input variable history shaft temperature data.

Preferably, the DCPU in the edge computing device of the side layer performs joint detection on multiple abnormal expression forms in the shaft temperature data by analyzing the shaft temperature residual sequence, and the specific processing procedure includes:

The DCPU of the edge computing device downloads the mapping relation between the pre-trained shaft temperature historical data and the current data from the vehicle layer TMC, inputs the historical shaft temperature data into the pre-established mapping relation between the shaft temperature historical data and the current data to obtain a shaft temperature predicted value at the current moment, and inputs the shaft temperature predicted value at the current momentTrue shaft temperature data y at current moment _t The difference results in a residual sequence E, expressed as:

E＝[e ₁ ,e ₂ ,…,e _t-1 ,e _t ] (2)

wherein e _t A residual error representing time t, which is expressed as:

secondly, in order to calculate the anomaly detection threshold values at different moments, the DCPU of the edge calculation device divides the residual sequence into mutually non-overlapping sub-sequences, the sub-sequence length being h, and for moment t, the residual sub-sequence being expressed as:

wherein h represents the number of residual values used for analysis; t is an integer multiple of h;

the DCPU of the edge computing device obtains the probability density function of each residual sub-sequence through a KDE methodFor a pair ofIntegrating to obtain a cumulative distribution function of each residual sub-sequence, which is expressed as:

the value of the sequence corresponding to the 0.05 quantile of the distribution is calculatedAs a lower limit, the value of +.A.A.was taken by the sequence corresponding to the 0.95 quantile>As an upper limit and as an anomaly detection non-parametric dynamic threshold, if the residual exceeds the anomaly detection non-parametric dynamic threshold, the data is identified as point anomalies, +. >Is->Is an inverse function of (2);

and (3) dividing the residual sequence into a plurality of sub-sequences with different lengths by taking the detected point abnormality as a demarcation point, applying an improved local outlier factor ILOF method to the residual sub-sequence by the DCPU of the edge computing device, and identifying the sub-sequence and the demarcation point thereof as a collective abnormality when the ILOF score is higher than an abnormality detection non-parameter dynamic threshold value, otherwise, identifying the sub-sequence as normal and the demarcation point of the sub-sequence as the point abnormality.

Preferably, the DCPU in the edge computing device performs a method for learning vibration signal characteristics and judging anomalies of the motor bearing, reflects the change of the working state of the traction transmission device through fractional dispersion entropy FrFDE, and reflects the change of the internal structure through the vibration signal, and the specific processing procedure includes:

the method comprises the steps of performing signal conditioning on electric charge quantity of an end layer acceleration sensor installed near a vibration source of equipment by adopting a charge amplifier, sending obtained acceleration detection quantity to a vehicle-mounted edge main control module of an edge layer through an internal bus interface module, storing the vehicle-mounted edge main control module in a storage array, performing real-time sliding window processing on data with window length of N and step length of L by using DCPU, and obtaining a one-dimensional vibration signal time sequence x= { x with length of N by sliding window each time ₁ ,x ₂ ,...,x _N Mapping the time series to y= { y using a normal cumulative distribution function ₁ ,y ₂ ,...,y _N }；

Wherein: gamma and sigma are the mean and standard deviation, respectively, of the time series x, y being mapped to {1,2,., c }, according to the following linear transformation:

wherein: round () is a rounding function; c is the number of categories;is the ith element in class c;

pair sequencePerforming phase space reconstruction with embedding dimension of m and delay factor of d to obtain new reconstruction vector +.>The following are provided:

each embedded vector is then insertedScattering pattern mapped to wave motion->

Assigned to each time sequenceThe number of all possible distribution patterns of (2 c-1) ^m-1 ；

For each possible spreading patternCalculate the relevant probability of its occurrence +.>The formula is as follows:

fractional order operators are introduced to capture more dynamic properties of the vibration signal, namely the Riemann-Liuvin operator:

D ^α f(x)＝D ⁿ I ^n-α f(x) (14)

wherein: i ^α Represents the alpha-order RL integral, D ^α For the alpha-order RL derivative, Γ (&) is the factorial function in real numbersAnd calculating fractional order fluctuation dispersion entropy under different alpha orders according to definition of shannon entropy by using a class of functions expanded on complex numbers:

preferably, the collaborative monitoring system is used for monitoring the vibration signal abnormally based on fractional order fluctuation dispersion entropy and accumulation and control diagram, and the specific processing procedure comprises:

The method comprises the steps that vibration signals of electric transmission equipment to be monitored are collected by using an acceleration sensor and a signal conditioning circuit of an end layer and are transmitted to an edge main control module of the side layer through an internal bus;

calculating fractional order fluctuation dispersion entropy of vibration signals under each sliding window in real time through DCPU in a main control module of the edge calculating device, tracking state change of equipment, and storing calculation results in a storage module;

the DCPU applies a cumulative and control graph CUSUM algorithm to track the track of the fractional order fluctuation dispersion entropy sequence, judges whether the monitoring data is abnormal by utilizing a threshold value, timely reports the detected abnormal point to a vehicle-mounted monitoring center when the equipment is abnormal, and reports the abnormal point to the CCU of a vehicle layer through a TCN after the vehicle-mounted monitoring center determines;

the expression of the CUSUM algorithm is:

x (t) is process detection data, g ⁺ And g ^- The cumulative sum of the positive and negative changes, respectively, the initial value of the parameter h being zero, i.e. g ⁺ (0)＝g ^- (0) Given a threshold T, if g ⁺ > T or g ^- The CUSUM algorithm detects anomalies and alerts the operator in time.

Preferably, the collaborative monitoring system is used for performing multi-source collaborative monitoring on the state of the motor according to the characteristics among different signals of multi-source data, and the specific processing procedure comprises:

Temperature X using the sensor module _temp Current X _c Torque X _tor Collecting, and transmitting the detection quantity to respective edge main control modules through the bus interface modules;

the DCPU in the edge main control module firstly carries out preliminary judgment on the data through FFT analysis or threshold value, if no abnormality exists, only carries out monitoring and display on the data, if the abnormality exists, carries out further fine processing on the data, and carries out stator temperature X of different sampling frequencies _temp Stator current X _c Torque signal X _tor Dividing the current data into equal-length data segments, determining the value of the length k of each data segment of the current data by comparing the diagnosis effect, determining the time t of the data segment according to the value k, and taking t for each segment of data duration of temperature and torque signals, so that the data segment can be determined to contain the number of data, and the data of three different signals input by the convolutional neural network are collected in the same time;

each data segment is normalized separately, where t=1, 2, …, k:

in DCPU in the edge master control module, for stator temperature X _temp Stator current X _c Torque signal X _tor The three different signals are subjected to feature extraction by adopting CNNs with different convolution operation parameters and the same other parameters, and the feature extraction flow is as follows:

Step 1, assuming that the input signal is x, performing convolution operation on the x:

y _i ^l ＝f(b _k ^l +conv1D(W ^l ,S _i ^l-1 )) (19)

where i=1, 2, …, C represents the number of convolution kernels,indicating the convolution result of the ith convolution kernel input x,is the output neuron of the layer 1 convolutional pooling module, < >>Represents the kth output offset, W, corresponding to the first convolution layer ^l Representing the weight of the layer l convolution layer, f ()'s activation function, conv1D represents one-dimensional convolution;

step 2, adopting maximum pooling to reduce dimension, and simultaneously carrying out batch normalization processing on output characteristics to obtain C characteristics N ^l ＝[n ₁ ^l ,n ₂ ^l ,...,n _C ^l ]The dimension of each feature is mx1, where max pool represents maximum pooling and BN represents batch normalization;

p _i ^l ＝maxpool(y _i ^l )

n _i ^l ＝BN(p _i ^l ) (20)

step 3 evaluate different feature importance using global average pooling and full connectivity layer, global_aveboost representing global average pooling, FC representing full connectivity layer, FC ² Representing C evaluation vectors, corresponding to C features respectively;

ga ^l ＝global_avepool(N ^l )

fc ¹ ＝FC(ga ^l ,C/r)

fc ² ＝FC(fc ¹ ,C) (21)

where r represents the dimension reduction ratio, C features are respectively associated with the evaluation vector fc ² Multiplying the C elements of (C) to obtain C weighted features y= [ Y ] ₁ ¹ ,y ₂ ¹ ,…,y _c ¹ ]；

In the respective edge computing device, 3 sets of m×1-dimensional features obtained for three convolutional neural networks are assumed to be F respectively ₁ 、F ₂ 、F ₃ ；

Will F ₁ 、F ₂ 、F ₃ Respectively transposed, and then connected by using a splicing function Conc, namely:

F＝Conc(Trans(F ₁ )、Trans(F ₂ )、Trans(F ₃ )) (23)

The edge main control module utilizes the Ethernet interface module to send the extracted characteristics to the vehicle-mounted monitoring center, the vehicle-mounted monitoring center fuses various characteristics, a Softmax function is used for carrying out diagnosis model training and model parameter determination, elements in an output vector are respectively mapped between (0 and 1), the sum of the elements is 1, the sum of the elements is used as the probability of each fault category, and the category with the largest probability is used as a final classification result;

wherein exp (z) _i ) J=1, 2, …, N represents the j-th input of the Softmax function, N is the total number of categories;

after the modeling of the vehicle-mounted monitoring center is completed, the model is transmitted to a corresponding edge computing device through an Ethernet interface module, the edge computing device transmits the fault to a central control unit through a TCN after fault diagnosis is carried out by using the model, the vehicle-mounted monitoring center sends relevant fault data to the central control unit through a vehicle gateway according to a collaborative monitoring result and a fault prediction result, and the central control unit adopts operations such as speed reduction, parking or switching on a traction system according to a preset program.

Preferably, the collaborative monitoring system is used for obtaining dynamic thresholds for a plurality of physical quantities through a collaborative learning algorithm, performing fault prediction and health management on monitoring equipment, determining different dynamic judgment thresholds according to the change of an operating environment, and obtaining an abnormal monitoring result.

According to the technical scheme provided by the embodiment of the invention, the invention provides a tailorable edge computing device for realizing collaborative monitoring and diagnosis for traction transmission and safety detection equipment of a motor train unit, which can be connected with sensors for detection of different types, and forms an edge computing device with unified framework and tailorable functions through an internal bus, a CPU unit, a storage unit and a memory, a communication unit and the like. Access, feature extraction and anomaly discovery of multiple types of sensors are achieved.

Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for the description of the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

Fig. 1 is a schematic logic structure diagram of an edge computing device according to an embodiment of the present invention.

Fig. 2 is a logic structure diagram of a sensor module in an edge computing device according to an embodiment of the present invention.

FIG. 3 is a schematic structural diagram of an edge master control module according to an embodiment of the present invention;

fig. 4 is a schematic diagram of a logic structure of a collaborative monitoring system of a train traction transmission device and a safety monitoring device according to an embodiment of the present invention;

fig. 5 is a schematic diagram of a task completed by each layer of the vehicle-side-end according to an embodiment of the present invention.

FIG. 6 is a block diagram of an LSTM circulation unit according to an embodiment of the present invention;

FIG. 7 is a schematic diagram of a normal shaft temperature data feature learning flow provided in an embodiment of the present invention;

FIG. 8 is a flowchart of an algorithm for intelligently detecting abnormal shaft temperature data according to an embodiment of the present invention;

FIG. 9 is a graph showing variation of fractional fluctuation entropy values at different orders α and different probabilities p according to an embodiment of the present invention;

fig. 10 is a flowchart of a multi-source collaborative monitoring method according to an embodiment of the present invention.

Detailed Description

Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary only for explaining the present invention and are not to be construed as limiting the present invention.

As used herein, the singular forms "a", "an", "the" and "the" are intended to include the plural forms as well, unless expressly stated otherwise, as understood by those skilled in the art. It will be further understood that the terms "comprises" and/or "comprising," when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof. It will be understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements may also be present. Further, "connected" or "coupled" as used herein may include wirelessly connected or coupled. The term "and/or" as used herein includes any and all combinations of one or more of the associated listed items.

It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the prior art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

For the purpose of facilitating an understanding of the embodiments of the invention, reference will now be made to the drawings of several specific embodiments illustrated in the drawings and in no way should be taken to limit the embodiments of the invention.

The invention relates to equipment applied to traction electric transmission systems and safety monitoring systems of high-speed railways, urban rail transit, intelligent trains and unmanned trains.

A schematic logic structure of an edge computing device according to an embodiment of the present invention is shown in FIG. 1. The edge computing device is a basis for realizing collaborative monitoring, the edge computing device can be connected with different numbers and types of sensors through circuit design, a central processing unit with powerful computing function is adopted, a larger storage space is configured, and more and unified network interfaces are arranged to realize data interaction among the edge computing devices, so that a collaborative monitoring system is formed. The edge computing device forms a collaborative monitoring system with unified architecture and tailorable functions through an internal bus, a CPU unit, a storage unit, a communication unit and the like, and realizes access, feature extraction and anomaly discovery of multiple types of sensors.

The edge computing device comprises a power supply module, a sensor module, a positioning module, an edge master control module, a storage module, an MVB interface module, an Ethernet interface module and a wireless interface module. The edge computing device adopts a modularized access mode and is connected with each other through an internal signal transmission bus, the internal signal transmission bus CAN be a 485 bus or a CAN bus, and an edge master control module CAN adopt a ModBus protocol when the 485 bus is utilized. The edge computing device can flexibly add or delete corresponding modules according to the characteristics of the monitoring equipment and the requirements of users, and is tailorable hardware.

The power module takes power from a vehicle-mounted power supply DC110V, obtains a direct current power supply required and suitable for each part of the edge computing device through DC-DC conversion, and supplies power to other modules through a power bus.

The positioning module adopts a GPS chip or a Beidou chip to provide position data for the edge computing device, and a singlechip is arranged in the positioning module and provides position data for the edge master control module through an internal signal transmission bus.

A schematic logic structure of a sensor module provided by the embodiment of the invention is shown in fig. 2. The sensor module mainly comprises a sensor, conditioning conversion, processing judgment, storage and communication. The sensor module adopts different sensors according to different monitoring devices, and the sensor types mainly comprise analog output, digital output or switching value output, 485 bus output and custom protocol output according to the requirements of monitoring objects and output signals. Different conditioning and converting circuits are adopted according to different output signals of the sensor, and the sensor for outputting analog quantity is provided with a conditioning circuit and an analog-to-digital conversion circuit and then is input to a CPU of the sensor module; the sensor of the digital quantity/switching quantity output is provided with a level conversion circuit and then is input to the CPU of the sensor module; the sensor output by the 485 bus is provided with a 485 interface circuit, and then is input to the CPU of the sensor module; the sensor output by the self-defined protocol is provided with a level conversion circuit or a CPU of a direct input value sensor module according to the characteristics of the output level. The CPU of the sensor module processes the sensor signals to obtain the detection quantity of the sensor, and the detection quantity is transmitted to the vehicle-mounted edge master control module through the internal signal transmission bus interface. The CPU of the sensor module can select the current commonly used 51-core singlechip, only the sensor data is subjected to preliminary pretreatment, and the storage capacity of the storage circuit can be small. The sensor module is close to the current monitoring system, can be deployed near the monitoring equipment according to actual conditions, can be deployed at the monitoring equipment together with the sensor according to monitoring tasks and deployment sites, can also be deployed at the monitoring equipment, and is deployed in the edge computing device.

Fig. 3 is a schematic structural logic diagram of an edge master control module according to an embodiment of the present invention, where the edge master control module includes a DCPU (Data processing Central Processing Unit ), an MCPU (Management Central Processing Unit, edge device management central processing unit), a data storage array, a Flash memory, a bus interface, and the like, as shown in fig. 3. After preprocessing the data of the sensor module, the DCPU acquires the characteristic quantity of the normal state and the state change quantity reflecting the state evolution by utilizing a characteristic extraction algorithm, and learns and models the normal data or the normal data characteristics to acquire the change rule of the normal state; judging whether the monitoring data is abnormal by utilizing a threshold value, so as to judge whether the equipment is abnormal; various intelligent algorithms and variable thresholds can also be utilized to determine outliers. The normal features and their descriptive model, and the outliers are stored in a data storage array. If the memory array of the edge master control module is full, the data is transmitted to the extended memory module for storage through the internal signal transmission bus. When the cooperative monitoring is needed, the DCPU performs data interaction with other edge master control modules through a unified network interface or interacts with a vehicle-mounted central control unit (CCU, central Control Unit) to obtain the needed monitoring data, performs cooperative anomaly judgment of multi-source data on the current monitoring data, and can realize dynamic threshold judgment or intelligent anomaly judgment. The MCPU of the edge master control module is used for managing the initialization of the edge computing devices, the management of internal bus transmission signals, the management of various communication, the management of various modules, the cooperative address allocation of different edge computing devices when the train is in initial operation, and the like. The DCPU and the MCPU exchange related data through the Flash memory. The DCPU and the MCPU perform control and data interaction with other modules of the edge computing device through an internal bus interface.

The memory module is an extended memory unit of the edge computing device, can be configured with a large number of memory chips, provides more memory space for normal running state and change thereof, abnormal points, multisource collaborative monitoring data characteristics and the like, is provided with a management CPU, a memory array and a bus interface, and performs data interaction with the edge master control module through an internal data transmission bus.

The MVB interface module is a main interface of the edge computing device connected with the TCN, and the data of the edge computing device are mostly transmitted through the vehicle-mounted Ethernet, so that the MVB interface module is used for receiving a control instruction of the vehicle-mounted CCU and reporting the key abnormal point to the CCU through the TCN.

The Ethernet interface module is a main communication network for connecting the edge computing devices with the TMC (Train Monitoring Center ), and the edge computing devices interact with each other through the Ethernet interface module.

The wireless interface module can enable one or more types of wireless communication such as ZigBee, NBiot or GPRS and the like to be set according to the need or not, and can be mainly used for receiving wireless sensor data possibly used later; the wireless interface module controls networking, address allocation and the like of the vehicle-mounted wireless sensor. The wireless interface module performs data transmission with the ground monitoring data center, and sends the processing result of the vehicle-mounted edge computing device to the ground monitoring center.

The edge computing device adopts modularization and is a flexible composition which can be cut, and different combinations or parameters can be set according to the type and characteristics of the monitoring sensor; according to the configuration requirement of the edge computing device, whether the wireless communication module is needed or not is set, and the tailorable and simplified setting mode of the edge computing device can meet the requirements of different carriages and monitoring equipment of a train.

The edge computing device is utilized to carry out cooperative monitoring on the multisource physical quantity of the traction transmission equipment and the safety monitoring equipment of the rail train, so as to form a cooperative monitoring system, as shown in fig. 4. And setting an edge computing device and a sensor module according to the characteristics of the monitoring equipment per train. The edge computing device is accessed into the TCN through the vehicle gateway by utilizing the MVB interface, is connected with the vehicle-mounted central control unit CCU, reports abnormality to the CCU and receives an instruction. The edge computing devices of each vehicle are interconnected by Ethernet and connected to TMC, which receives the relevant data and issues results and tasks. TMC can be set in existing systems as required, or can be reset according to train monitoring tasks.

The collaborative monitoring system of the traction transmission and safety detection equipment of the rail train of the embodiment of the invention can be described by three layers, and a schematic diagram of a work task completed by each layer of the train-side-end provided by the embodiment of the invention is shown in fig. 5. The end layer mainly refers to a sensor module and is arranged at the train equipment. The sensor module selects the types and the quantity of the sensors according to the characteristics of the physical quantity of the monitored equipment, performs data acquisition on the state of the monitored equipment, performs preprocessing on the data, performs threshold judgment and early warning, and sends the judgment result to the edge computing device. The sensor module of the end layer also receives instructions of the edge computing device, including a determination threshold and a dynamic threshold of abnormality judgment, setting of data acquisition frequency and output transmission quantity, and the like.

The edge layer is an edge computing device. The edge computing device is provided with two cars, and is used for respectively receiving the voltage and the current of a traction motor arranged on the bogie, and the temperature, the vibration, the speed and other data of the bearings of the traction motor stator and the bearing gearbox. The edge computing device may have access to one or more sensor modules, as needed for condition monitoring. The edge computing device performs aggregation processing, feature extraction of a normal operation rule and an evolution rule of the data transmitted by the sensor module, and abnormal state discrimination and analysis based on the normal operation rule and the evolution rule of the normal operation rule, performs fault diagnosis on the transmitted abnormal point, and determines whether the transmitted abnormal point is the determined abnormal point. The edge computing device stores the analysis processing result into the storage module; and determining data to be transmitted to the train monitoring center according to the train monitoring center of the train layer. The edge computing node receives control instructions sent by the train monitoring center of the train and comprises fault diagnosis results, confirmation of abnormal judgment thresholds, characteristics of normal state rules, management of storage capacity and storage modules and the like.

The train layer is a train monitoring center TMC. The TMC has the same composition as the edge computing device, and is connected to the edge computing device of each car through Ethernet and manages the edge computing device of the whole train. The TMC receives monitoring and analysis results of the same physical quantity from different edge computing devices, compares and judges the monitoring and analysis results, and carries out intelligent learning and mining model training and parameter determination of the same physical quantity on the basis; receiving monitoring and analysis results of different physical quantities monitored in a cooperative state, comparing and judging, and carrying out model training and parameter determination by utilizing intelligent learning and excavation of a plurality of physical quantities on the basis; determining an abnormality discrimination threshold and a dynamic threshold of different rail physical quantities, carrying out fault prediction and health management of equipment according to the results of an edge computing device of a vehicle, evaluating the state of the equipment and estimating the service life, and providing a primary maintenance scheme to adapt to the requirements of intelligent trains and unmanned trains on a monitoring system. The TMC performs data interaction with the ground monitoring center through a mobile communication network such as wireless GPRS (general packet radio service ) or 5G. The TMC receives instructions of the ground monitoring center, including train fault prediction results, abnormal judgment standards, cooperative model structures and parameters, monitoring model structures and parameters, state monitoring tasks required by the ground monitoring center and the like.

The collaborative monitoring system is constructed, so that the same monitoring equipment can be monitored simultaneously by multiple physical quantities, the state of the monitoring equipment can be judged by fusing the multiple physical quantities, and the normal state and state evolution, equipment degradation and possible equipment faults are judged. The reliability of the current rail train state monitoring result is improved, the PHM requirement is met, and the intelligent train and the unmanned train are adapted to the requirement of the monitoring system. And constructing a collaborative monitoring system, and simultaneously, realizing a collaborative monitoring method in an edge computing device. The temperature monitoring of the traction motor bearing and the temperature of the gearbox bearing are taken as an example for illustration.

The application process of the cooperative monitoring system of the rail train electric transmission equipment and the safety monitoring equipment provided by the embodiment of the invention comprises the following steps:

1: and (5) normal shaft temperature data characteristic learning.

The characteristic learning process of the normal shaft temperature data is completed in the car layer TMC, and the data set required by model training is from historical normal shaft temperature data stored by the TMC or normal shaft temperature data transmitted by a boundary layer edge computing device. The feature learning algorithm adopts an LSTM (long short-term memory) model, and fig. 6 is a structure diagram of an LSTM circulation unit provided by the embodiment of the invention. LSTM introduces a new internal state c in each cell _t Carry out linear cyclic information transfer and output information to the external state h of the hidden layer _t . By introducing forgetting door f _t Input gate i _t And transportGo out o _t Controlling information quantity and raising LSTM memory ability.

Fig. 7 is a schematic diagram of a normal shaft temperature data feature learning flow provided in an embodiment of the present invention, and a specific processing procedure includes: firstly, normalizing normal shaft temperature data, and secondly, carrying out sliding window operation on the normal shaft temperature data to enable the normal shaft temperature data to adapt to the input of an LSTM model. And inputting the preprocessed shaft temperature sequence into an LSTM model for training, and finally obtaining the mapping relation between the shaft temperature historical data and the current data. The characteristic learning process is completed based on historical normal shaft temperature data stored by the vehicle layer TMC or normal shaft temperature data transmitted by the side layer edge computing device and by utilizing the computing performance of the vehicle layer TMC. The final result of the shaft temperature data characteristic learning is a black box model reflecting shaft temperature historical data and current data, namely:

temp _t ＝f(temp _t-1 ,temp _t-2 ,…,temp _t-m ) (1)

in the formula, temp _t The shaft temperature at time t; f (-) represents a nonlinear mapping relation, and the invention refers to a black box model without an explicit expression; m is the length of the model input variable (i.e., the historical shaft temperature data).

2: and (5) detecting abnormal shaft temperature data intelligently.

Fig. 8 is a flowchart of an intelligent detection algorithm for abnormal shaft temperature data, which is provided in an embodiment of the present invention and runs on a DCPU of an edge computing device of an edge layer. The key of abnormal shaft temperature data detection is to perform joint detection on various abnormal expression forms in the shaft temperature data by analyzing a shaft temperature residual sequence. The shaft temperature residual sequence refers to the difference between the current shaft temperature data and the current time shaft temperature prediction data obtained through the mapping relation between the pre-established shaft temperature historical data and the current data; the plurality of abnormal manifestations refers to point abnormalities and subsequence abnormalities that are commonly represented in the shaft temperature data. The general flow of the abnormal shaft temperature data intelligent detection is as follows: the DCPU of the edge computing device first downloads the mapping relationship between the pre-trained shaft temperature history data and the current data from the car layer TMC, and then performs a diagnostic process in the DCPU of the edge computing device.

First, the DCPU of the edge computing device downloads pre-trained from the car-layer TMCThe mapping relation between the shaft temperature historical data and the current data is obtained by inputting the historical shaft temperature data into the pre-established mapping relation between the shaft temperature historical data and the current data, and the shaft temperature predicted value at the current moment is obtained True shaft temperature data y at current moment _t The difference results in a residual sequence E, which can be expressed as:

E＝[e ₁ ,e ₂ ,…,e _t-1 ,e _t ] (2)

wherein e _t A residual error representing time t, which is expressed as

Secondly, in order to calculate the abnormal detection threshold values at different moments, the DCPU of the edge calculation device divides the residual sequence into sub-sequences which are not overlapped with each other, and the sub-sequence length is h. For time t, the residual sub-sequence may be expressed as:

wherein h represents the number of residual values used for analysis; t is an integer multiple of h. The DCPU of the edge computing device executes KDE (kernel density estimation ) method to each residual sub-sequence to carry out non-parameter fitting to the distribution, takes the 0.95 quantile and the 0.05 quantile of the calculated distribution as the upper limit and the lower limit, and takes the 0.95 quantile and the 0.05 quantile as the non-parameter dynamic threshold of anomaly detection. The KDE method is a non-parametric probability density estimation method that is applicable to the problem of modeling probability density functions (probability density function, PDF) that distribute unknown data.

The probability density function of each residual sub-sequence can be obtained by a KDE methodDCPU pair of edge computing device>Integration is performed to obtain a cumulative distribution function (Cumulative Distribution Function, CDF) for each residual sub-sequence, which can be expressed as:

The value of the sequence corresponding to the 0.05 quantile of the distribution is calculatedAs a lower limit, the value of +.A.A.was taken by the sequence corresponding to the 0.95 quantile>As an upper limit and as an anomaly detection non-parametric dynamic threshold. If the residual exceeds the threshold, the data is identified as a point anomaly. />Is->Is an inverse function of (c).

Finally, the residual sequence is subdivided into a plurality of sub-sequences of different lengths (the sub-sequences do not include demarcation points) using the detected point anomalies as demarcation points. The DCPU of the edge computing device applies a modified local outlier (improved local outlier factor, ILOF) method to the residual sub-sequence, which sub-sequence and its demarcation point are identified as collectively abnormal when the ILOF score is above a threshold, whereas the sub-sequence is normal and the demarcation point of the sub-sequence is still identified as point abnormal. Notably, when the residual subsequence contains fewer data points (less is defined as less than 3 data points by the present invention) or no data points, the subsequence and its two-terminal demarcation points are directly determined to be a collective anomaly, without calculating an ILOF score.

The ILOF algorithm is an improvement on the traditional LOF algorithm, is a method based on density information, and can describe the deviation degree of data and adjacent data, and can be used as an anomaly score to identify anomaly data. Unlike the conventional LOF algorithm, the ILOF algorithm replaces the Euclidean distance or Manhattan distance in the conventional LOF algorithm with a Soft-dynamic time warping (Soft dynamic time warping, soft-DTW) distance to solve the problem of dimensional differences between the residual sub-sequences of the shaft temperature to be analyzed. The ILOF algorithm widens the application range of the LOF algorithm, and breaks through the limitation that the traditional LOF algorithm can only be used for isomorphic data.

The goal of Soft-DTW is to find the optimal alignment between two heterogeneous sequences to overcome the limitation of data isomerism on distance computation. Specifically, a given sequence a= [ a ] ₁ ,a ₂ ,...,a _n ]And B= [ B ] ₁ ,b ₂ ,…,b _m ]The lengths are n and m, respectively. Defining an n m distance matrixWherein: dist (dist) _i,j ＝||a _i -b _j || ₂ ,i＝1,2,...n,j＝1,2,…m。

The Soft-DTW distance can be expressed as:

in the method, in the process of the invention,the minimum distance returned for Soft-DTW, which is calculated by the following recursion:

wherein:

thus, the Soft-DTW distance is replaced by the Euclidean distance or the Manhattan distanceThe ILOF algorithm is performed to calculate the ILOF score for the sub-sequence. When the ILOF score of a sub-sequence is greater than a given threshold θ _ILOF In this case, the subsequence is determined to be abnormal. In the present invention, the threshold value θ _ILOF Setting to three times the standard deviation of all sub-sequences ILOF scores.

3: method for learning vibration signal characteristics and judging abnormality of motor bearing

The invention provides a new and effective characteristic extraction index based on fluctuation dispersion entropy (Fluctuation dispersion entropy, FDE), namely fractional dispersion entropy (Fractional fluctuation dispersion entropy, frFDE) reflects the change of the working state of traction transmission equipment, and the change of an internal structure is reflected by a vibration signal. The specific characteristic extraction flow of the fractional order fluctuation dispersion entropy is as follows:

And step 1, carrying out signal conditioning on the charge quantity of an end layer acceleration sensor arranged near the equipment vibration source by adopting a charge amplifier, and sending the obtained acceleration detection quantity to a vehicle-mounted edge master control module of an edge layer through an internal bus interface module and storing the acceleration detection quantity in a storage array. Meanwhile, the DCPU carries out real-time sliding window processing with the window length of N and the step length of L on the data, and the one-dimensional vibration signal time sequence x= { x with the length of N is obtained by sliding the window each time ₁ ,x ₂ ,...,x _N -mapping the time series to y= { y using a normal cumulative distribution function (normal cumulative distribution function, NCDF) ₁ ,y ₂ ,...,y _N }；

/>

Wherein: gamma and sigma are the mean and standard deviation, respectively, of the time series x. Then y is mapped to {1, 2..c } according to the following linear transformation:

wherein: round () is a rounding function; c is the number of categories;is the ith element in class c.

Step 2: pair sequencePerforming phase space reconstruction with embedding dimension of m and delay factor of d to obtain new reconstruction vector +.>The following are provided:

each embedded vector is then insertedScattering pattern mapped to wave motion->

Assigned to each time sequenceThe number of all possible distribution patterns of (2 c-1) ^m-1 。

Step 3: for each possible spreading patternCalculate the relevant probability of its occurrence +. >The formula is as follows:

step 4: fractional order operators are introduced to capture more dynamic properties of the vibration signal, namely the Riemann-Liuvin operator:

D ^α f(x)＝D ⁿ I ^n-α f(x) (14)

wherein: i ^α Represents the alpha-order RL integral, D ^α For the α -order RL derivative, Γ (&) is a class of functions where the factorial function extends over real and complex numbers. According to the definition of shannon entropy, calculating fractional order fluctuation dispersion entropy under different alpha orders:

from the above flow, the fractional order fluctuation dispersion entropy has the following characteristics:

1) The state characteristics of the vibration signals under different orders alpha can be extracted, the fractional fluctuation dispersion entropy change curves under different orders alpha and different probabilities p are shown in fig. 9, particularly when the orders are larger than 0 and smaller than 1, the p entropy values under different probabilities have great difference, the vibration signals are more sensitive to weak state evolution of the signals, and different states of equipment can be better distinguished.

2) The monitoring quantity is the fractional order fluctuation dispersion entropy of the vibration signal instead of the original signal, so that the reliability of the monitoring result is improved.

3) The FrFDE algorithm has smaller time complexity and space complexity, reduces the calculation load and is very suitable for real-time signal processing of an edge layer.

4. Monitoring vibration signals, and obtaining abnormal monitoring results through an intelligent algorithm

The design idea of the accumulation and control map (cumulative sum control chart, CUSUM) algorithm is to accumulate and sum the small offsets in the process to achieve an amplification effect, thereby improving the sensitivity of the anomaly monitoring process to the small offsets. The CUSUM expression is:

x (t) is process detection data, g ⁺ And g ^- Respectively the cumulative sum of the positive and negative changes. In order to avoid detecting abnormal changes without actual changes or slow drift, the algorithm also relies on a parameter h for drift correction. Their initial value is zero, i.e. g ⁺ (0)＝g ^- (0) =0. Given a threshold T, if g ⁺ > T or g ^- The CUSUM algorithm can detect anomalies and alert the operator in time.

Vibration signal anomaly detection flow based on fractional order fluctuation dispersion entropy and accumulation and control diagram is as follows:

step 1: the method comprises the steps that vibration signals of electric transmission equipment to be monitored are collected by using an acceleration sensor and a signal conditioning circuit of an end layer and are transmitted to an edge main control module of the side layer through an internal bus;

step 2: calculating fractional order fluctuation dispersion entropy of the vibration signal under each sliding window in real time through DCPU in the main control module, tracking state change of equipment, and storing calculation results in the storage module;

Step 3: the DCPU uses the accumulation and control diagram to track the track of the fractional fluctuation dispersion entropy sequence, judges whether the monitoring data is abnormal by utilizing a threshold value, reports the detected abnormal point to the vehicle-mounted monitoring center in time when the equipment is abnormal, and reports the abnormal point to the CCU of the vehicle layer through the TCN after the vehicle-mounted monitoring center determines.

5: multi-source collaborative monitoring method for motor state

The single physical quantity monitoring and judging are sometimes inaccurate, the characteristics among different signals of the multi-source data are adopted in the edge computing device, weights are distributed according to the importance degree of the characteristics, and the accuracy and the reliability of the equipment fault diagnosis result are improved. The multi-source collaborative monitoring method comprises feature extraction, feature splicing and fault diagnosis of different signals, and fig. 10 is a flowchart of the multi-source collaborative monitoring method according to an embodiment of the present invention, and the specific processing procedure includes:

(1) Feature extraction of different signals

First, the sensor module is used for measuring the temperature X _temp Current X _c Torque X _tor Collecting and transmitting the detection quantity to the respective edge master control modules through the bus interface modules. The DCPU at the edge main control module firstly carries out preliminary judgment on the data through FFT analysis or threshold value, if no abnormality exists, only carries out monitoring and display on the data, if the abnormality exists, carries out further fine processing on the data, and carries out stator temperature X at different sampling frequencies _temp Stator current X _c Torque signal X _tor The current sampling frequency is highest in order to ensure the sampling precision, so that the value of the length k of each data segment of the current data can be determined by comparing the diagnosis effect. And determining the time t of the data segment according to the k value, and taking t for each segment of data duration of the temperature and torque signals, so as to determine the number of data contained in the data segment. The data of three different signals input by the convolutional neural network are collected at the same time.

Each data segment is normalized separately, where t=1, 2, …, k:

for stator temperature X _temp Stator current X _c Torque signal X _tor And extracting the characteristics of three different signals by adopting CNNs with different convolution operation parameters and the same other parameters. The feature extraction flow is as follows:

step 1, assuming that the input signal is x, performing convolution operation on the input signal:

y _i ^l ＝f(b _k ^l +conv1D(W ^l ,S _i ^l-1 )) (19)

where i=1, 2, …, C represents the number of convolution kernels,indicating the convolution result of the ith convolution kernel input x,is the output neuron of the layer 1 convolutional pooling module, < >>Represents the kth output offset, W, corresponding to the first convolution layer ^l Representing the weights of the layer l convolution layer, f () activation functions, conv1D represents one-dimensional convolution.

Step 2, adopting maximum pooling to reduce dimension, and simultaneously carrying out batch normalization processing on output characteristics to obtain C characteristics N ^l ＝[n ₁ ^l ,n ₂ ^l ,…，n _C ^l ]The dimension of each feature is mx1, where max pool represents maximum pooling and BN represents batch normalization.

p _i ^l ＝max pool(y _i ^l )

n _i ^l ＝BN(p _i ^l ) (20)

Step 3, the global average pooling and full connection layer are utilized to evaluate different feature importance. Wherein global_aveboost represents global average pooling, FC represents fully connected layer, FC ² And C evaluation vectors are represented and correspond to C characteristics respectively.

ga ^l ＝global_avepool(N ^l )

fc ¹ ＝FC(ga ^l ,C/r)

fc ² ＝FC(fc ¹ ,C) (21)

Where r represents the dimension reduction ratio, C features are respectively associated with the evaluation vector fc ² The features will be weighted for importance by multiplying the C elements in (C). Resulting in C weighted features y= [ Y ] ₁ ¹ ,y ₂ ¹ ,…,y _c ¹ ]。

(2) Feature stitching in edge computing devices

In the respective edge computing device, 3 sets of m×1-dimensional features obtained for three convolutional neural networks are assumed to be F respectively ₁ 、F ₂ 、F ₃ 。

Will F ₁ 、F ₂ 、F ₃ The transposition (Trans) is performed respectively, and then the concatenation function Conc is used for connection, namely:

F＝Conc(Trans(F ₁ )、Trans(F ₂ )、Trans(F ₃ )) (23)

(3) Fault classification identification

The extracted features are stored and backed up in an edge device, the extracted features are sent to a vehicle-mounted monitoring center by utilizing an Ethernet interface module, various features are fused in the vehicle-mounted monitoring center, a collaborative diagnosis model is built for training by combining normal data, parameters are determined, and a threshold value and a dynamic threshold value are determined. When abnormal data exists, the abnormal data is fused and judged by adopting a model, and the abnormal data is uploaded to a ground monitoring center through a vehicle-mounted GPRS. The determined diagnosis model and parameters are sent to the corresponding edge computing device through the Ethernet interface module. The determined faults are also transmitted to a corresponding edge computing device through the Ethernet interface module, the edge computing device transmits the faults to a central control unit through TCN, and the central control unit performs operations such as speed reduction, stopping or switching on the traction system according to a preset program.

The modeling principle of the vehicle-mounted monitoring center is as follows, in the full-connection layer of the convolutional neural network, a Softmax function is used for fault classification, elements in an output vector can be respectively mapped between (0 and 1), and the sum of the elements is 1, so that the elements can be regarded as the probability of each fault class, and the class with the largest probability is regarded as a final classification result.

Wherein exp (z) _i ) J=1, 2, …, N represents the j-th input of the Softmax function, N is the total number of categories.

6. The collaborative monitoring method comprises the following steps: the collaborative monitoring system based on the edge computing device comprises a new monitoring method, mainly comprising the following steps:

(1) monitoring a certain physical quantity, realizing feature extraction of normal running state and state evolution feature extraction thereof

(2) Monitoring a certain physical quantity, and obtaining an abnormal monitoring result through an intelligent algorithm

(3) The dynamic threshold value is obtained for a plurality of physical quantities through a collaborative learning algorithm, different dynamic judgment threshold values can be determined according to the change of the running environment, and abnormal judgment is carried out, so that the problems of high false alarm rate and high false alarm rate of the current monitoring system are solved.

7. Dynamic threshold

Most of the abnormality detection processes in the existing vehicle-mounted safety equipment are based on a threshold method, for example, the abnormality detection of a shaft temperature signal and a vibration signal adopts the threshold method to judge data abnormality, and the abnormality detection process cannot adapt to the actual operation working condition or health state of the equipment in many cases. The method is integrated with a rule-based method in the abnormality detection method based on the threshold value, so that the abnormality detection threshold value is changed into a dynamic threshold value considering the running condition and the health state of the equipment.

The rule-based method in the present invention aims to dynamically adjust the abnormality detection threshold by a rule set in advance. Specifically, when considering the actual operating conditions of the train equipment: 1) For the axle temperature threshold setting, if the traction motor current exceeds the rated current, the train runs at full load, and each component of the running part of the train cannot bear higher running temperature, so that the axle temperature threshold needs to be adjusted downwards to forecast the axle temperature fault in advance; 2) For the setting of the vibration threshold, if the train speed exceeds 350km/h, a small deviation of the vibration signal will be indicative of a train running gear failure, so that the vibration threshold needs to be adjusted down at this time. When considering the health of train equipment: if the equipment failure rate is low or the equipment operates healthily for a long time, the failure threshold value can be adjusted upwards; if it is determined that the device failure rate is high or the lifetime is approaching a safety boundary, the threshold is adjusted down.

The operation state of the electric drive is evaluated by the cooperative monitoring of a plurality of physical quantities.

In summary, according to the embodiment of the invention, by adding the embedded edge computing device and the storage medium to each type of sensor and performing data processing and feature extraction on the sensor end, the feature quantity is transmitted only through the train communication network, so that the data transmission quantity is reduced, and the deployment quantity of the sensors can be greatly increased.

The edge computing device performs data processing and analysis near the sensor end, extracts characteristic data, transmits the characteristic data and abnormal data, and greatly reduces data transmission quantity. The sensor deployment quantity can be increased, and the requirements of intelligent trains and unmanned trains on state monitoring are met. Substantial features and significant advances.

Each type of sensor is added with embedded edge computing equipment and storage media, and can store normal state characteristics of train running lines and climate conditions and characteristic evolution rules thereof.

Each type of sensor is added with an embedded edge computing device and a storage medium, so that normal state characteristics of a train running line and weather conditions can be stored; the edge computing device has a self-learning function, can extract and store the characteristics of the normal state of the train operation process, and can obtain state evolution. The train detection device does not have the function, which is a substantial characteristic and significant progress.

Each type of sensor is added with an embedded edge computing device and a storage medium, an intelligent learning algorithm is realized by using the edge computing device, different thresholds can be set under different conditions, and the accuracy is improved.

By adopting the edge computing device, the dynamic threshold value can be obtained by utilizing the learning algorithm on the basis of recording the normal state characteristics and the evolution characteristics, so that the false alarm rate is reduced. The edge computing device is capable of evaluating the sensor state to distinguish between anomalies caused by the sensor state and anomalies caused by the device state.

The traction electric transmission system state fusion judgment is realized through the vehicle edge computing device, and the equipment state can be judged more accurately than the independent system monitoring and judgment.

By adopting the edge computing device, the fusion judgment is carried out on the equipment state at the feature layer, so that the possible abnormality and the accurate estimation of the residual life of the equipment which cannot be obtained by the independent monitoring and the threshold judgment at present can be found. The train detection device does not have the function, which is a substantial characteristic and significant progress.

Those of ordinary skill in the art will appreciate that: the drawing is a schematic diagram of one embodiment and the modules or flows in the drawing are not necessarily required to practice the invention.

From the above description of embodiments, it will be apparent to those skilled in the art that the present invention may be implemented in software plus a necessary general hardware platform. Based on such understanding, the technical solution of the present invention may be embodied essentially or in a part contributing to the prior art in the form of a software product, which may be stored in a storage medium, such as a ROM/RAM, a magnetic disk, an optical disk, etc., including several instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the method described in the embodiments or some parts of the embodiments of the present invention.

In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments. In particular, for apparatus or system embodiments, since they are substantially similar to method embodiments, the description is relatively simple, with reference to the description of method embodiments in part. The apparatus and system embodiments described above are merely illustrative, wherein the elements illustrated as separate elements may or may not be physically separate, and the elements shown as elements may or may not be physical elements, may be located in one place, or may be distributed over a plurality of network elements. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art will understand and implement the present invention without undue burden.

The present invention is not limited to the above-mentioned embodiments, and any changes or substitutions that can be easily understood by those skilled in the art within the technical scope of the present invention are intended to be included in the scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims

1. An edge computing device, comprising: the device comprises a power module, a sensor module, an edge master control module and a storage module which are mutually connected through an internal signal transmission bus;

the sensor module comprises a sensor, conditioning conversion, processing judgment, storage and communication, wherein different conditioning and conversion circuits are adopted according to different output signals of the sensor, the sensor signals are preprocessed and the threshold value is judged to obtain normal state quantity and/or abnormal quantity of a monitored object, and the normal state quantity and/or abnormal quantity of the monitored object is sent to the edge master control module through an internal signal transmission bus interface;

the edge master control module comprises a data processing central processing unit (DCPU), an edge device Management Central Processing Unit (MCPU), a data storage array, a Flash memory and a bus interface, wherein the DCPU gathers data of one or more sensor modules, acquires characteristic quantity of a normal state and state variation quantity reflecting evolution of the normal state by utilizing a characteristic extraction algorithm for the normal state quantity, learns and models normal data or normal data characteristics, and obtains a variation rule of the normal state; analyzing and further judging the abnormal quantity, combining the normal data characteristics and the evolving state change rate, performing fault diagnosis on the abnormal quantity, and determining a threshold revision quantity for frequent faults according to the fault diagnosis result; the MCPU realizes the initialization of the edge computing device, the internal bus transmission, the management of various communication modules and various modules, and the cooperative address allocation of different edge computing devices when the train is in initial operation; the DCPU and the MCPU exchange related data through a Flash memory, and control and data exchange are carried out between the DCPU and other modules of the edge computing device through an internal bus interface;

2. The apparatus of claim 1, wherein said apparatus further comprises: the system comprises a positioning module, a multifunctional communication bus MVB interface module, an Ethernet interface module and a wireless interface module;

3. A collaborative monitoring system for traction electric drive and safety monitoring equipment based on an edge computing device according to claim 1 or 2, comprising: an edge computing device and a sensor module are arranged in each carriage of the rail train and are used for carrying out multi-source physical quantity collaborative monitoring on traction electric transmission equipment and safety monitoring equipment, the edge computing devices of each rail train are connected with each other through Ethernet and are connected to a train monitoring center TMC, the TMC receives related data, carries out model training and parameter determination, determines various thresholds, carries out fault prediction and health management and issues results and tasks;

4. The system of claim 3, wherein the collaborative monitoring system comprises a car floor, an edge floor, and an end floor;

5. The system of claim 4, wherein the collaborative monitoring system is configured to perform normal shaft temperature data feature learning by using a long-short-term memory network LSTM model through a vehicle layer TMC, and establish a mapping relationship between shaft temperature history data and current data through a deep learning algorithm, and the specific processing procedure includes:

temp _t ＝f(temp _t-1 ,temp _t-2 ,…,temp _t-m ) (1)

6. The system of claim 5, wherein the DCPU in the edge computing device of the edge layer performs joint detection on multiple abnormal manifestations in the shaft temperature data by analyzing the shaft temperature residual sequence, and the specific processing procedure includes:

E＝[e ₁ ,e ₂ ,…,e _t-1 ,e _t ] (2)

wherein e _t A residual error representing time t, which is expressed as:

the DCPU of the edge computing device obtains the probability density function of each residual sub-sequence through a KDE methodFor->Integrating to obtain a cumulative distribution function of each residual sub-sequence, which is expressed as:

7. The system of claim 4, wherein the DCPU in the edge computing device performs a method for learning and anomaly discrimination of vibration signal characteristics of the motor bearing, reflects a change in an operating state of the traction drive device by fractional dispersion entropy FrFDE, and reflects a change in an internal structure by the vibration signal, and the specific processing procedure includes:

each embedded vector is then insertedScattering pattern mapped to wave motion->

D ^α f(x)＝D ⁿ I ^n-α f(x) (14)

wherein: i ^α Represents the alpha-order RL integral, D ^α For the alpha-order RL derivative, Γ (&) is a class of functions of which the factorial function extends over real numbers and complex numbers, and fractional order fluctuation dispersion entropy under different alpha orders is calculated according to the definition of shannon entropy:

8. the system of claim 4, wherein the collaborative monitoring system is configured to perform anomaly monitoring on the vibration signal based on fractional order wave dispersion entropy and accumulation and control charts, the specific process comprising:

the expression of the CUSUM algorithm is:

9. The system of claim 4, wherein the collaborative monitoring system is configured to perform multi-source collaborative monitoring of motor status based on characteristics between different signals of multi-source data, and the specific processing procedure includes:

each data segment is normalized separately, where t=1, 2, …, k:

y _i ^l ＝f(b _k ^l +conv1D(W ^l ,S _i ^l-1 )) (19)

where i=1, 2, …, C represents the number of convolution kernels,indicating the convolution result of the ith convolution check input x,/->Is the output neuron of the layer 1 convolutional pooling module, < >>Represents the kth output offset, W, corresponding to the first convolution layer ^l Representing the weight of the layer l convolution layer, f ()'s activation function, conv1D represents one-dimensional convolution;

step 2, adopting maximum pooling to reduce dimension, and simultaneously carrying out batch normalization processing on output characteristics to obtain C characteristics N ^l ＝[n ₁ ^l ,n ₂ ^l ,…,n _C ^l ]The dimension of each feature is mx1, where max pool represents maximum pooling and BN represents batch normalization;

p _i ^l ＝max pool(y _i ^l )

n _i ^l ＝BN(p _i ^l ) (20)

step 3, evaluating different feature importance by using global average pooling and full connection layer, wherein global_aveboost represents global average pooling and FC represents FCFull tie layer fc ² Representing C evaluation vectors, corresponding to C features respectively;

ga ^l ＝global_avepool(N ^l )

fc ¹ ＝FC(ga ^l ,C/r)

fc ² ＝FC(fc ¹ ,C) (21)

F＝Conc(Trans(F ₁ )、Trans(F ₂ )、Trans(F ₃ )) (23)

10. The system of claim 4, wherein the collaborative monitoring system is configured to obtain dynamic thresholds for a plurality of physical quantities through a collaborative learning algorithm, perform fault prediction and health management on the monitoring device, determine different dynamic determination thresholds according to a change of an operating environment, and obtain an abnormal monitoring result.