CN112216085A - Equipment key load-bearing structural member health monitoring system based on edge calculation and intelligent identification of updated samples - Google Patents
Equipment key load-bearing structural member health monitoring system based on edge calculation and intelligent identification of updated samples Download PDFInfo
- Publication number
- CN112216085A CN112216085A CN202010966981.0A CN202010966981A CN112216085A CN 112216085 A CN112216085 A CN 112216085A CN 202010966981 A CN202010966981 A CN 202010966981A CN 112216085 A CN112216085 A CN 112216085A
- Authority
- CN
- China
- Prior art keywords
- edge
- fault diagnosis
- cloud computing
- computing layer
- layer
- 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.)
- Granted
Links
- 238000012544 monitoring process Methods 0.000 title claims abstract description 51
- 238000004364 calculation method Methods 0.000 title claims abstract description 32
- 230000036541 health Effects 0.000 title claims abstract description 23
- 238000003745 diagnosis Methods 0.000 claims abstract description 50
- 238000012549 training Methods 0.000 claims abstract description 24
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 17
- 230000005540 biological transmission Effects 0.000 claims abstract description 13
- 230000004927 fusion Effects 0.000 claims abstract description 11
- 238000000605 extraction Methods 0.000 claims abstract description 8
- 239000008186 active pharmaceutical agent Substances 0.000 claims abstract description 7
- 230000009467 reduction Effects 0.000 claims abstract description 7
- 238000000034 method Methods 0.000 claims description 20
- 239000011159 matrix material Substances 0.000 claims description 17
- 230000008859 change Effects 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 10
- 238000012545 processing Methods 0.000 claims description 10
- QVFWZNCVPCJQOP-UHFFFAOYSA-N chloralodol Chemical compound CC(O)(C)CC(C)OC(O)C(Cl)(Cl)Cl QVFWZNCVPCJQOP-UHFFFAOYSA-N 0.000 claims description 8
- 239000013598 vector Substances 0.000 claims description 8
- 230000006835 compression Effects 0.000 claims description 7
- 238000007906 compression Methods 0.000 claims description 7
- 230000001360 synchronised effect Effects 0.000 claims description 7
- 238000005070 sampling Methods 0.000 claims description 4
- 238000003860 storage Methods 0.000 claims description 4
- 101100172132 Mus musculus Eif3a gene Proteins 0.000 claims description 3
- 230000004044 response Effects 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 claims description 2
- 230000002708 enhancing effect Effects 0.000 claims description 2
- 230000000737 periodic effect Effects 0.000 claims description 2
- 238000000926 separation method Methods 0.000 claims description 2
- 208000034155 RAS-associated autoimmune leukoproliferative disease Diseases 0.000 claims 2
- 201000009200 autoimmune lymphoproliferative syndrome type 4 Diseases 0.000 claims 2
- 238000001514 detection method Methods 0.000 claims 1
- 230000010354 integration Effects 0.000 abstract description 5
- 238000010276 construction Methods 0.000 abstract description 4
- 238000005094 computer simulation Methods 0.000 abstract description 3
- 230000004931 aggregating effect Effects 0.000 abstract 1
- 230000008569 process Effects 0.000 description 9
- 238000001228 spectrum Methods 0.000 description 4
- 238000012937 correction Methods 0.000 description 3
- 238000004088 simulation Methods 0.000 description 3
- 239000000725 suspension Substances 0.000 description 3
- 238000012360 testing method Methods 0.000 description 3
- 230000001133 acceleration Effects 0.000 description 2
- 230000008901 benefit Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 230000004913 activation Effects 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000003111 delayed effect Effects 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000009792 diffusion process Methods 0.000 description 1
- 238000006073 displacement reaction Methods 0.000 description 1
- 230000009977 dual effect Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 239000000284 extract Substances 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 230000008092 positive effect Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 238000004451 qualitative analysis Methods 0.000 description 1
- 238000011158 quantitative evaluation Methods 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 238000005096 rolling process Methods 0.000 description 1
- 238000005728 strengthening Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
- 238000012795 verification Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G08—SIGNALLING
- G08C—TRANSMISSION SYSTEMS FOR MEASURED VALUES, CONTROL OR SIMILAR SIGNALS
- G08C17/00—Arrangements for transmitting signals characterised by the use of a wireless electrical link
- G08C17/02—Arrangements for transmitting signals characterised by the use of a wireless electrical link using a radio link
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01M—TESTING STATIC OR DYNAMIC BALANCE OF MACHINES OR STRUCTURES; TESTING OF STRUCTURES OR APPARATUS, NOT OTHERWISE PROVIDED FOR
- G01M13/00—Testing of machine parts
-
- G—PHYSICS
- G08—SIGNALLING
- G08C—TRANSMISSION SYSTEMS FOR MEASURED VALUES, CONTROL OR SIMILAR SIGNALS
- G08C19/00—Electric signal transmission systems
Abstract
The invention discloses a health monitoring system for a key load-bearing structural member of equipment based on edge calculation and online update sample intelligent identification. The system comprises a center cloud computing layer, an edge cloud computing layer and an edge end computing layer. The central cloud computing layer receives and stores data from the edge cloud computing layer, so that real-time monitoring and early warning of all safety state information entering the cloud are realized; the training of the off-line fault diagnosis model is realized through the establishment of a dynamic simulation model, the enhancement of weak signal faults, the feature extraction and dimension reduction, the construction of feature subsets and the construction and integration of sub-models. The edge cloud computing layer distributes computing tasks to optimal edge computing nodes through a task scheduling algorithm to achieve task scheduling of updating the fault diagnosis model; and aggregating the data and the fault diagnosis result of the edge terminal calculation layer through the local area network to realize the priority scheduling of data transmission and the real-time monitoring of the data acquired by the sensor. The edge end computing layer evaluates the safety condition nearby by deploying a fault diagnosis model on the edge computing node, and realizes online monitoring of the fault; and identifying an updating sample through an evidence weighting fusion algorithm based on the DS, so as to realize the updating of the fault diagnosis model.
Description
Technical Field
The invention relates to the field of health monitoring of equipment key bearing structural members, in particular to a health monitoring system of the equipment key bearing structural members based on edge calculation and intelligent identification of updated samples.
Background
At present, the health monitoring of the equipment key load-bearing structural member mainly establishes a complex fault recognition model through a computer technology, and carries out online monitoring of qualitative diagnosis of faults by taking state data of the equipment key load-bearing structural member as input, however, the online monitoring mode has certain disadvantages: from the perspective of identification accuracy, when a key load-bearing structural member of a high-speed train or the like runs under extreme service conditions such as alternating load, high temperature, high pressure, variable speed, vibration, humidity and the like for a long time, concept drift of a monitored object can be caused, statistical characteristics shown by faults change in an unpredictable manner, and prediction accuracy of a fault diagnosis model trained offline is further reduced. In order to ensure the performance of a fault diagnosis and prediction model, modeling data is required to cover all working conditions, model parameters can adapt to the change of the working conditions, and the model needs to have a correction function; considering from the perspective of server load pressure, along with the increase of monitored key bearing structural members, the load capacity of a server needs to be continuously expanded, which can cause the problems of increase of operation cost, reduction of cloud platform processing performance and the like; considering from the aspect of real-time processing requirements, once a key force-bearing structural member has a safety problem, a result is diagnosed at the nearest node in the first time, and a safety strategy is implemented in time, and if a large amount of data still needs to be queued in sequence and transmitted to a cloud platform for processing through a network link, the load, performance and bandwidth pressure of a server are increased, and meanwhile, the diagnosis and pre-evaluation of serious safety problems are greatly delayed.
Disclosure of Invention
The invention provides a key load-bearing structural member health monitoring system based on edge calculation and online sample updating intelligent identification, which integrates an edge calculation technology into the key load-bearing structural member health monitoring system, provides a mutually connected calculation and storage resource architecture, and utilizes near-end edge calculation node resources, thereby effectively reducing the transmission quantity of network data and improving the response speed and the prediction precision. The accuracy of fault diagnosis is improved by using the offline training model and the online updating model, and meanwhile, the timeliness of model updating is guaranteed by the task scheduling algorithm of the edge computing layer.
The invention is realized by the following technical scheme: a health monitoring system based on edge calculation and online update sample intelligent identification and provided with key bearing structural parts comprises: the device comprises a center cloud computing layer, an edge cloud computing layer and an edge end computing layer.
The central cloud computing layer is mainly responsible for receiving, storing and analyzing data, and providing real-time monitoring and early warning of all safety state information entering the cloud and training of an offline fault diagnosis model.
The off-line fault diagnosis model training comprises five steps of establishing a dynamic simulation model, enhancing weak fault signals, extracting and reducing the dimension of features, establishing feature subsets, and establishing and integrating a sub-model, wherein the sub-model selects the RVFL network.
The edge cloud computing layer realizes the updating task scheduling, the data transmission priority scheduling and the real-time monitoring of the fault diagnosis model, and the method specifically comprises the following steps:
(1) fault diagnosis model update task scheduling
And after any edge computing node identifies the updating sample, the edge cloud computing layer distributes an updating model task to the edge computing node through a polling task scheduling algorithm.
(2) Data transmission priority scheduling
During task scheduling, the edge cloud computing layer gathers data and fault diagnosis results of the edge end computing layer through an LoRa protocol, and uploads the data to the center cloud computing layer through 4G transparent transmission; due to the limitation of the computing performance and the network bandwidth of the loRa concentrator, when relevant data of the edge computing layer are gathered and congestion occurs, the edge cloud computing layer is communicated with the loRa concentrator in a CSMA/CA mode, and the loRa concentrator is combined with the characteristic of channel idle monitoring, so that the problem of node channel conflict is solved, and the accuracy and the reliability of data reporting are improved.
(3) Real-time monitoring
Carrying out real-time overall monitoring on data acquired by the edge end calculation layer, and popping up an alarm frame on a monitoring interface in time to send out an alarm bell once alarm information exists; the safety state of each current compartment of the whole vehicle can be displayed through the forms of a real-time curve, a historical curve and the like, and the real-time global change trend is monitored.
The edge end computing layer realizes online fault monitoring and online fault diagnosis model updating, and the method specifically comprises the following steps:
(1) on-line monitoring
The edge computing node collects vibration signal data of the key bearing structural member through an RS485 protocol; further, the fault is diagnosed nearby by utilizing a fault diagnosis model deployed on the edge computing node.
(2) Online update
And the edge computing node identifies an update sample, the update sample identification considers the influence of the RALD value and the RPE value of the new sample subjected to weak signal enhancement, the RALD and the RPE value are fused and output by using an improved weighted evidence fusion algorithm of the DS evidence distance, and the fault diagnosis model is updated after the update sample is identified.
Compared with the prior art, the invention has the advantages and positive effects that:
according to the scheme, part of computing tasks which are born by an original cloud computing center and have high requirements on bandwidth, computing resources and instantaneity are migrated to the edge computing nodes near the data source for execution, and the transmission quantity of network data is reduced and the timeliness of fault diagnosis is improved by utilizing the near-end edge computing node resources; an improved weighted evidence fusion algorithm based on DS evidence distance is provided, an updated sample is identified and output for fusion of RALD and RPE values, fault diagnosis errors caused by concept drift are effectively reduced through online updating of a fault diagnosis model, and fault diagnosis precision is improved; and the updating task of the fault diagnosis model is distributed to different edge computing nodes through a task scheduling strategy, so that the real-time performance of fault diagnosis is further improved.
Drawings
Fig. 1 is a structural frame diagram of a key force-bearing structural member health monitoring system based on edge calculation and intelligent identification of an update sample according to an embodiment of the invention;
FIG. 2 is a flowchart illustrating an off-line model training process according to an embodiment of the present invention;
fig. 3 is a flowchart of online monitoring and online updating according to an embodiment of the present invention.
Detailed Description
In order to make the objects, features and advantages of the present invention more clearly understood, the present invention will be further described with reference to the accompanying drawings and examples. In the following description, numerous specific details are set forth in order to provide a thorough understanding of the present invention, however, the present invention may be practiced in other ways than those described herein, and thus, the present invention is not limited to the specific embodiments disclosed below.
As shown in fig. 1, the embodiment discloses a high-speed railway bogie health monitoring system based on edge computing and intelligent identification of updated samples, which distributes resources such as computing, storage, bandwidth and the like to edge computing nodes, and is divided into a center cloud computing layer, an edge cloud computing layer and an edge end computing layer.
1. Central cloud computing layer
And the central cloud computing layer is used for monitoring and early warning safety state information and training an offline fault diagnosis model.
(1) Monitoring and early warning of safety state information
The working state, the communication channel state and the system storage space use condition of the acceleration sensor are monitored in real time, data uploaded by the edge cloud computing layer are stored and analyzed, and monitoring and early warning of all high-speed rail real-time safety state information entering the cloud are achieved.
(2) Offline model training
As shown in fig. 2, the central cloud computing layer performs offline training on a fault diagnosis model, and a sub-model in the fault diagnosis model selects an RVFL network.
1) Establishment of dynamic simulation model
The training data is obtained by establishing an integrated model of a dynamic system by a method for assembling vehicle subsystems by adopting ADAMS/Rail. The bogie system consists of a front bogie and a rear bogie, and the main parameters are basically consistent. During modeling, all subsystems are assumed to be rigid bodies and have 6 degrees of freedom.
A steering frame system creating step:
a) hard points are defined to create wheel sets. And establishing a double-wheel pair according to the coordinate position, the vehicle attribute and the like.
b) The architecture is built based on dual wheel pairs, since the wheel pair locations are already fixed, at which point hard spots need not be created.
c) Two front axle boxes and two rear axle boxes are sequentially established at the axle box positions according to the vehicle attributes.
d) The method comprises the steps of establishing a suspension system, firstly establishing structural frame points, establishing primary suspension according to the structural frame points, then establishing an assembly object for connecting a vehicle body and a bogie, and finally establishing the structural frame points and establishing secondary suspension.
e) And (3) establishing a vertical damper, a transverse damper and an anti-snaking damper according to the attribute of the vehicle body, then establishing a stop gear, a connecting pair between the axle box and the wheel pair, and preferably establishing an anti-rolling torsion bar and a rotating arm elastic node, namely completing the establishment of a front steering frame system model.
f) And according to the distance between the established front bogie system and the front bogie and the rear bogie, the rear bogie system can be established.
g) Simulating a vibration signal acquired by a bogie acceleration sensor by using a health state model, obtaining a deviation by combining the health state vibration signal measured by an experiment, and constructing a model error correction matrix;
h) and correcting the simulated fault sample by using the model error correction matrix, and taking the fault simulation sample as a training sample.
100 fault simulation samples are generated for each fault type (10 fault types in total), wherein 60 samples are used as a training set, 20 samples are used as a test set, 20 samples are used as a verification set, the length of each simulation fault sample is 0.5s, the sampling frequency is 20000Hz, and therefore the number of sampling points is 10000.
2) Weak fault signal enhancement
The strengthening and effective extraction of weak fault features under strong noise are the basis for establishing a fault quantitative evaluation model. The vibration signal acquired under strong noise needs to be input as a model after fault feature enhancement and extraction. The correlation kurtosis is taken as a target function, a filter with finite impulse response is selected, the correlation kurtosis value of a known signal is maximized through an iterative convergence criterion, the periodic characteristics of sudden change components in the vibration signal are fully considered, the continuous pulse sudden change components covered by strong noise in the vibration signal are highlighted through deconvolution operation of the signal, and weak fault signal separation is realized.
3) Feature extraction and dimensionality reduction
And performing time-frequency analysis on the separated signals by adopting synchronous compression wavelet transform, and acquiring a time-frequency matrix capable of representing different faults after discrete wavelet transform, phase transformation, synchronous compression and signal reconstruction processing. The problem of overhigh dimension of the fault signal feature space after feature extraction still exists, the non-negative matrix is further utilized to decompose, simplify and optimize the feature space, and the sparse expression matrix obtained through calculation is used as the final signal feature.
a) Firstly, discrete wavelet transform is carried out on the enhanced fault signal f (t):
wherein Wf(a, b) represents the wavelet coefficient spectrum, #*The complex conjugate of the mother wavelet function is represented, a represents the scale parameter, and b represents the displacement parameter.
b) Energy diffusion often occurs in the actually obtained wavelet coefficient spectrum, so that the time frequency spectrum becomes fuzzy, but the phase is not influenced by scale change, so that the instantaneous frequency is calculated by using the phase obtained by the wavelet coefficient:
c) (a, b) and (w) can be established using the calculated instantaneous frequencyf(a, b) and b), synchronous compression transformation, and then, redistributing the energy of the time-scale plane to convert the energy into a time-frequency plane:
since the synchronous compression transform rearranges the complex spectrum of the wavelet transform only along the frequency axis direction, so that the synchronous compression transform is reversible, the reconstructed signal can be expressed as:
wherein f isk(tm) Represents the valid signal extracted by the k-th channel,
the representation depends on the function of the selected mother wavelet, Re represents the real part, lk(tm) Representing the effective signal band range, the present embodiment extracts effective signals of 50 channels.
Each sample can be converted to a time-frequency matrix of 50 x 10000:
the problem of overhigh dimension of the fault signal feature space after feature extraction still exists, the non-negative matrix is further utilized to decompose, simplify and optimize the feature space, and the sparse expression matrix obtained through calculation is used as the final signal feature. Detailed optimization procedure
The following were used:
normalizing the time-frequency matrixes corresponding to all the data samples;
a) converting the time-frequency matrix into a column vector;
b) randomly selecting 60 samples of each fault type to form a training matrix Vn×mWhere n is the sample dimension, 500000 in this example, and m is the number of samples, 600 in this example.
c) Solving an optimization problem by taking Kullback-Leber divergence as an objective function:
d) finally, a base matrix W is obtainedn×rSum coefficient matrix Hr×mAt this time Hr×mI.e. representing the feature matrix in the new feature space, r is selected to be 100, and finally the feature vector dimension of each sample is reduced to 100. The feature matrices of the test set and validation set may be obtained using the same computational approach.
4) Building feature subsets
And (3) establishing a feature subset of a training set by adopting a BootStrap algorithm with back-put uniform sampling according to the feature vector after dimensionality reduction:
where J is the number of training subsets, i.e. the number of candidate submodels; l represents the number of feature vectors, and the total number is k. In this example, J is selected to be 10 and k is selected to be 60.
5) Construction and integration of submodels
The RVFL network has good nonlinear fitting capability and rapid modeling potential, and is particularly suitable for distributed environments.
Furthermore, the RVFL network is linear in parameters, can obtain the optimal parameters by using a standard linear regression method, and can be effectively realized in low-cost hardware (such as a sensor or a mobile device). The RVFL network is therefore selected as a sub-model for fault qualitative analysis.
The construction and combination steps of the submodels are as follows:
a) and constructing the RVFL network with a three-layer structure with the node number of 100-200-10, wherein x is an n-dimensional input variable, y is an expected output variable, and the sigmoid function is selected as the activation function h of the hidden layer. Can be expressed as:
before learning begins, selecting omega and b independent of training data according to experience as weights and biases from an input layer to a hidden layer, and obtaining a hidden layer kernel mapping matrix H for calculating an output layer part:
b) calculating an output weight beta from a hidden layer to an output layer according to a standard regularization least square principle, wherein the output weight beta comprises the following steps:
the weight β can be found as:
β＝(HTH+λI)-1HTY
y is a column vector consisting of Y corresponding to x in the training sample space; λ is a constant; i denotes an identity matrix.
c) The output of the RVFL model is as follows:
where L represents the number of hidden layer nodes.
d) For the integration submodel based onCalculating weight coefficient by AWF algorithm, and adopting
The forms are combined.
2. Edge cloud computing layer
The edge cloud computing layer realizes the task scheduling of updating the fault diagnosis model and the priority scheduling of data transmission through a scheduling strategy, and monitors the data collected by the edge end computing layer in real time.
(1) Fault diagnosis model update task scheduling
After any edge computing node identifies an update sample, the edge cloud computing layer distributes a fault diagnosis model update task through a polling scheduling algorithm, and the task scheduling process comprises the following 3 steps:
a) the parallelism is set by a contrast test, which is set to 10 in the embodiment, and the parallelism is divided into the component parallelism and the topology parallelism, wherein the former represents the number of threads, and the latter represents the number of processes of the JVM virtual machine. The processes in charge of distributing tasks and monitoring the running state of the system on the edge computing main node distribute the tasks layer by layer according to the parallelism setting and scheduling strategy.
b) Distributing the threads to JVM virtual machine processes installed on the edge computing node, wherein one or more threads can run on one JVM virtual machine process, and a plurality of distributed tasks can run on the one JVM virtual machine process; the scheduling algorithm adopts a polling mechanism, task slots are introduced to realize the allocation of computing resources, the number of the task slots of one edge computing node represents the number of the processes of the virtual machine which can be born, and each task slot has certain CPU resources and memory resources according to the configuration. When a thread needs to be allocated, the edge computing nodes are at least sorted according to the number of the available task slots, then the first available task slot is taken out from each edge computing node in sequence, and the thread is allocated in sequence. If there are unallocated threads after a round, then the next available task slot is taken from each edge compute node, and so on until all threads have been allocated.
c) Virtual machine processes are assigned to edge compute nodes.
(2) Data transmission priority scheduling
The number of data and fault diagnosis results of the edge end calculation layer is large, the types are complex, the positions are uncertain, and the data and the fault diagnosis results are scattered in each carriage.
Therefore, if a proper transmission strategy is not adopted, a plurality of acquisition nodes report data to the routing node at the same time, and a channel collision phenomenon is caused. Aiming at the existing problems, the edge cloud computing layer gathers data and fault diagnosis results of the edge end computing layer through an LoRa protocol, and uploads the data to the center cloud computing layer through 4G transparent transmission; because of the limit of the computing performance of the LoRa concentrator and the network bandwidth, the edge computing nodes communicate with the edge cloud computing layer in a CSMA/CA mode, whether a channel is used or not is detected, and if the channel is detected to be idle, data is sent out after waiting for a random time. If the receiving end receives the frame correctly, after a period of time interval, the receiving end sends an acknowledgement frame ACK to the sending end. And the sending end receives the ACK frame, determines that the data is correctly transmitted, and sends the data after a period of time interval. By adopting the communication mode and combining the characteristic that LoRa has channel idle monitoring, the problem of node channel conflict is solved, and the accuracy and reliability of data reporting are improved.
(3) Real-time monitoring
Carrying out real-time overall monitoring on data acquired by the edge end calculation layer, and popping up an alarm frame on a monitoring interface in time to send out an alarm bell once alarm information exists; the safety state of each current compartment of the whole vehicle can be displayed through the forms of a real-time curve, a historical curve and the like, and the real-time global change trend is monitored.
3. Edge side computation layer
As shown in fig. 3, the edge computing layer implements online fault monitoring and online fault diagnosis model updating.
(1) Fault on-line monitoring
The sensor node collects vibration signal data of the high-speed rail bogie through an RS485 protocol. Further, a fault diagnosis model deployed on the edge computing node is utilized to diagnose the fault nearby;
1) micrometrizing a New sampleWeak signal enhancement/feature extraction and dimension reduction processing to obtain features
Then based on the j old integration submodel, the output is predicted:
wherein
Representing the jth old integration submodel.
2) And calculating the weight coefficient of the integrated sub-model by adopting an online AWF algorithm:
3) The online prediction output for the new sample is:
(2) online update
And the edge calculation node identifies an update sample, the update sample identification considers the influence of an RALD value and an RPE value of the new sample subjected to weak signal enhancement, the RALD and the RPE respectively represent a relative approximate linear dependence value of the new sample facing the old modeling sample base, and a relative prediction error value of the new sample based on the old model. And fusing and outputting the RALD and the RPE values by using an improved weighted evidence fusion algorithm of the DS evidence distance, and updating the fault diagnosis model after identifying the updated sample.
1) First calculate xk+1ALD Absolute values of a library of relatively modeled samples
Wherein
Representing the old set of modeling sample data.
2) The Relative ALD (RALD) value a of the new sample is then calculatedk+1:
Wherein
Representing the ALD value of the ith sample relative to all other k-1 samples in the modeled sample library.
3) The Relative Prediction Error (RPE) at time k +1 is then calculated:
wherein
And representing the relative difference between the predicted value and the true value of the ith sample of the modeling sample library.
4) Fused output of the RALD value and the RPE value: because the quality of the obtained evidence is divided into good and bad due to the difference between the processing precision and the anti-interference capability of the evidence source during evidence fusion, an improved weighted evidence fusion algorithm based on DS evidence distance is adopted to perform weighted average processing on original evidences (RALD and RPE) to obtain a reference evidence for comparison, then the distance between each evidence and the reference object is calculated, and the calculation amount is further reduced, and the specific steps are as follows:
a) computing the original evidence piWeight coefficient w ofiThe formula is as follows
Wherein p is1Is RALD, p2Is RPE, EiFor each evidence piEntropy of evidence of (1).
b) Preprocessing the evidence and calculating the weighted average value p of the evidenceWAE。
c) Calculating piAnd pWAEJousselme distance between:
d) calculate each piece of evidence piAnd pWAESimilarity of (2):
Sim(pi,pWAE)＝1-d(pi,pWAE)
e) normalizing the similarity between the evidences in the previous step to obtain an evidence ciReliability of (2):
f) confidence of evidence Crd (p)i) Carrying out weighted average calculation on the original evidence by using the weight coefficient as the evidence;
g) carrying out n-1 times of iterative fusion on the weighted averaged evidence according to a DS fusion rule to obtain Usk+1。
5) Using gravity center method to align Usk+1And performing deblurring processing.
Wherein, thetacomIs a sample selection threshold, determined according to expert system as 1.5, Fthre() As a function of the threshold. When S isk+1When 1, the new sample is identified as the updated sample.
6) And finally, updating the fault diagnosis model by using the same method as the off-line model training of the central cloud computing layer through the updating task scheduling of the edge cloud computing layer model.
The above description is only a preferred embodiment of the present invention, and not intended to limit the present invention in other forms, and any person skilled in the art may apply the above modifications or changes to the equivalent embodiments with equivalent changes, without departing from the technical spirit of the present invention, and any simple modification, equivalent change and change made to the above embodiments according to the technical spirit of the present invention still belong to the protection scope of the technical spirit of the present invention.
Claims (10)
1. The utility model provides an equip key load structure health monitoring system based on edge calculation and online update sample intelligent recognition which characterized in that, with resources such as calculation, storage, bandwidth share to the edge calculation node, the system includes: the device comprises a center cloud computing layer, an edge cloud computing layer and an edge end computing layer.
The central cloud computing layer is used for realizing real-time monitoring and early warning of all safety state information entering the cloud and training of an offline fault diagnosis model.
The edge cloud computing layer realizes updating task scheduling, data transmission priority scheduling and real-time monitoring of the edge computing node fault diagnosis model.
And the edge end computing layer realizes online fault monitoring and online fault diagnosis model updating.
2. The equipment key force-bearing structural member health monitoring system based on edge calculation and intelligent identification of updated samples as claimed in claim 1, wherein: the method for enhancing the weak fault signals in the off-line fault diagnosis model training of the central cloud computing layer comprises the following steps:
the correlation kurtosis is taken as a target function, a filter with finite impulse response is selected, the correlation kurtosis value of a known signal is maximized through an iterative convergence criterion, the periodic characteristics of sudden change components in the vibration signal are fully considered, the continuous pulse sudden change components covered by strong noise in the vibration signal are highlighted through deconvolution operation of the signal, and weak fault signal separation is realized.
3. The equipment key force-bearing structural member health detection system based on edge calculation and online update sample intelligent identification as claimed in claim 1, wherein: the feature extraction and dimension reduction in the off-line fault diagnosis model training of the central cloud computing layer adopt the following modes:
performing time-frequency analysis on the separated signals by adopting synchronous compression wavelet transform, and acquiring time-frequency matrixes capable of representing different faults after discrete wavelet transform, phase transformation, synchronous compression and signal reconstruction processing; and further decomposing, simplifying and optimizing a feature space by using a non-negative matrix, and taking the computed sparse expression matrix as a final signal feature.
4. The equipment key force-bearing structural member health monitoring system based on edge calculation and intelligent identification of updated samples as claimed in claim 1, wherein: the method for constructing the feature subset in the offline fault diagnosis model training of the central cloud computing layer comprises the following steps:
and (3) establishing a feature subset of a training set by adopting a BootStrap algorithm with back-put uniform sampling according to the feature vector after dimensionality reduction:
where J is the number of training subsets, i.e. the number of candidate submodels; l represents the number of feature vectors, and the total number is k.
5. The equipment key force-bearing structural member health monitoring system based on edge calculation and intelligent identification of updated samples as claimed in claim 1, wherein: the method for constructing and combining the neutron model in the off-line fault diagnosis model training of the central cloud computing layer comprises the following steps:
(1) constructing a three-layer random vector function connection (RVFL) network;
(2) calculating the weight from the hidden layer to the output layer according to a standard regularization least square principle;
(3) and calculating and combining weight coefficients of the integrated sub-models based on an AWF algorithm.
6. The equipment key force-bearing structural member health monitoring system based on edge calculation and intelligent identification of updated samples as claimed in claim 1, wherein: the fault diagnosis model updating task scheduling in the edge cloud computing layer adopts the following mode:
and after any edge computing node identifies the updating sample, the edge cloud computing layer distributes an updating model task to the edge computing node through a polling task scheduling algorithm.
7. The equipment key force-bearing structural member health monitoring system based on edge calculation and intelligent identification of updated samples as claimed in claim 1, wherein: the data transmission priority scheduling in the edge cloud computing layer adopts the following mode:
while updating the model, the edge cloud computing layer gathers data and fault diagnosis results of the edge end computing layer through an LoRa protocol, and uploads the data to the center cloud computing layer through 4G transparent transmission; and the edge computing nodes communicate with the edge cloud computing layer in a CSMA/CA mode.
8. The equipment key force-bearing structural member health monitoring system based on edge calculation and intelligent identification of updated samples as claimed in claim 1, wherein: the real-time monitoring in the edge cloud computing layer adopts the following mode:
carrying out real-time overall monitoring on data acquired by the edge end calculation layer, and popping up an alarm frame on a monitoring interface in time to send out an alarm bell once alarm information exists; the safety state of each current compartment of the whole vehicle can be displayed through the forms of a real-time curve, a historical curve and the like, and the real-time global change trend is monitored.
9. The equipment key force-bearing structural member health monitoring system based on edge calculation and intelligent identification of updated samples as claimed in claim 1, wherein: the fault on-line monitoring in the edge terminal computing layer adopts the following mode:
and deploying the fault diagnosis model at the edge computing node, and acquiring sensor data by the edge computing node through an RS485 protocol to input the data into the fault diagnosis model so as to evaluate and diagnose the safety condition nearby.
10. The equipment key force-bearing structural member health monitoring system based on edge calculation and intelligent identification of updated samples as claimed in claim 1, wherein: the online updating of the fault diagnosis model in the edge end computing layer adopts the following mode:
and (3) the updated sample identification considers the influence of the RALD value and the RPE value of the new sample subjected to weak signal enhancement, the RALD value and the RPE value are subjected to fusion output by using an improved weighted evidence fusion algorithm of DS evidence distance, then the center-of-gravity method is adopted for deblurring processing, the updated sample is judged according to a threshold value, and the fault diagnosis model is updated after the updated sample is identified.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010966981.0A CN112216085B (en) | 2020-09-15 | 2020-09-15 | Equipment key load-bearing structural member health monitoring system based on edge calculation and online update sample intelligent identification |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010966981.0A CN112216085B (en) | 2020-09-15 | 2020-09-15 | Equipment key load-bearing structural member health monitoring system based on edge calculation and online update sample intelligent identification |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN112216085A true CN112216085A (en) | 2021-01-12 |
| CN112216085B CN112216085B (en) | 2022-05-10 |
Family
ID=74049559
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010966981.0A Active CN112216085B (en) | 2020-09-15 | 2020-09-15 | Equipment key load-bearing structural member health monitoring system based on edge calculation and online update sample intelligent identification |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN112216085B (en) |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113255584A (en) * | 2021-06-22 | 2021-08-13 | 德明通讯（上海）股份有限公司 | Fault diagnosis and monitoring system based on edge calculation |
| CN113507491A (en) * | 2021-04-30 | 2021-10-15 | 华中农业大学 | Method and system for uploading all-chain information of clean egg production in real time |
| CN113973029A (en) * | 2021-10-27 | 2022-01-25 | 广东电网有限责任公司江门供电局 | Low-voltage Internet of things edge computing intelligent gateway system and communication method |
| CN114057053A (en) * | 2022-01-18 | 2022-02-18 | 杭州浅水数字技术有限公司 | Method for monitoring fatigue degree of component of special machine |
| CN115946132A (en) * | 2023-03-15 | 2023-04-11 | 江西丹巴赫机器人股份有限公司 | Intelligent dense warehouse and logistics robot system and fault monitoring method thereof |
| CN116467570A (en) * | 2023-06-16 | 2023-07-21 | 西南交通大学 | Heavy-duty locomotive coupler swing angle quantitative identification method based on data driving |
Citations (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040176879A1 (en) * | 2003-03-07 | 2004-09-09 | Menon Sunil K. | Transient fault detection system and method using Hidden Markov Models |
| CN102067526A (en) * | 2008-07-03 | 2011-05-18 | 中兴通讯股份有限公司 | Synchronization, scheduling, network management and frequency assignment method of a layered wireless access system |
| US20120185728A1 (en) * | 2010-12-24 | 2012-07-19 | Commonwealth Scientific And Industrial Research Organisation | System and method for detecting and/or diagnosing faults in multi-variable systems |
| CN102928877A (en) * | 2012-11-14 | 2013-02-13 | 中国石油集团川庆钻探工程有限公司地球物理勘探公司 | Seismic property combination method based on Dempster/Shafe (D-S) evidence theory |
| CN103150454A (en) * | 2013-03-27 | 2013-06-12 | 山东大学 | Dynamic machine learning modeling method based on sample recommending and labeling |
| CN103632363A (en) * | 2013-08-27 | 2014-03-12 | 河海大学 | Object-level high-resolution remote sensing image change detection method based on multi-scale fusion |
| CN106404055A (en) * | 2016-08-29 | 2017-02-15 | 孟玲 | Power transmission network tower monitoring system |
| EP3181427A2 (en) * | 2015-12-18 | 2017-06-21 | General Electric Company | Sensor signal processing system and method |
| CN109406117A (en) * | 2018-09-28 | 2019-03-01 | 东南大学 | A kind of multichannel remote oscillation signal monitor system and monitoring method based on improved RBFNN algorithm |
| CN109491301A (en) * | 2019-01-23 | 2019-03-19 | 东莞固高自动化技术有限公司 | Industry internet intelligent controller based on edge calculations architectural framework |
| CN109547585A (en) * | 2019-01-14 | 2019-03-29 | 中国雄安集团数字城市科技有限公司 | A method of based on being switched fast for edge calculations cloud |
| CN109933004A (en) * | 2019-03-27 | 2019-06-25 | 苏芯物联技术(南京)有限公司 | The machine failure diagnosis and prediction method and system cooperateed with based on edge calculations and cloud |
| CN110059601A (en) * | 2019-04-10 | 2019-07-26 | 西安交通大学 | A kind of multi-feature extraction and the intelligent failure diagnosis method merged |
| CN110394688A (en) * | 2019-09-02 | 2019-11-01 | 太原科技大学 | Conditions of machine tool monitoring method based on edge calculations |
| CN110646183A (en) * | 2019-09-04 | 2020-01-03 | 国网浙江省电力有限公司嘉兴供电公司 | High-voltage circuit breaker mechanical property quick detection device based on thing networking |
| CN111259927A (en) * | 2020-01-08 | 2020-06-09 | 西北工业大学 | Rocket engine fault diagnosis method based on neural network and evidence theory |
| US20200209109A1 (en) * | 2018-12-28 | 2020-07-02 | Shanghai United Imaging Intelligence Co., Ltd. | Systems and methods for fault diagnosis |
| CN111366816A (en) * | 2020-04-26 | 2020-07-03 | 华北电力大学 | Power grid fault diagnosis method based on machine learning |
| CN111442926A (en) * | 2020-01-11 | 2020-07-24 | 哈尔滨理工大学 | Fault diagnosis method for rolling bearings of different models under variable load based on deep characteristic migration |
| CN111651312A (en) * | 2020-07-28 | 2020-09-11 | 南京宏声科技有限公司 | Three-level bidirectional early warning method and system for equipment fault based on edge calculation |
-
2020
- 2020-09-15 CN CN202010966981.0A patent/CN112216085B/en active Active
Patent Citations (20)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20040176879A1 (en) * | 2003-03-07 | 2004-09-09 | Menon Sunil K. | Transient fault detection system and method using Hidden Markov Models |
| CN102067526A (en) * | 2008-07-03 | 2011-05-18 | 中兴通讯股份有限公司 | Synchronization, scheduling, network management and frequency assignment method of a layered wireless access system |
| US20120185728A1 (en) * | 2010-12-24 | 2012-07-19 | Commonwealth Scientific And Industrial Research Organisation | System and method for detecting and/or diagnosing faults in multi-variable systems |
| CN102928877A (en) * | 2012-11-14 | 2013-02-13 | 中国石油集团川庆钻探工程有限公司地球物理勘探公司 | Seismic property combination method based on Dempster/Shafe (D-S) evidence theory |
| CN103150454A (en) * | 2013-03-27 | 2013-06-12 | 山东大学 | Dynamic machine learning modeling method based on sample recommending and labeling |
| CN103632363A (en) * | 2013-08-27 | 2014-03-12 | 河海大学 | Object-level high-resolution remote sensing image change detection method based on multi-scale fusion |
| EP3181427A2 (en) * | 2015-12-18 | 2017-06-21 | General Electric Company | Sensor signal processing system and method |
| CN106404055A (en) * | 2016-08-29 | 2017-02-15 | 孟玲 | Power transmission network tower monitoring system |
| CN109406117A (en) * | 2018-09-28 | 2019-03-01 | 东南大学 | A kind of multichannel remote oscillation signal monitor system and monitoring method based on improved RBFNN algorithm |
| US20200209109A1 (en) * | 2018-12-28 | 2020-07-02 | Shanghai United Imaging Intelligence Co., Ltd. | Systems and methods for fault diagnosis |
| CN109547585A (en) * | 2019-01-14 | 2019-03-29 | 中国雄安集团数字城市科技有限公司 | A method of based on being switched fast for edge calculations cloud |
| CN109491301A (en) * | 2019-01-23 | 2019-03-19 | 东莞固高自动化技术有限公司 | Industry internet intelligent controller based on edge calculations architectural framework |
| CN109933004A (en) * | 2019-03-27 | 2019-06-25 | 苏芯物联技术(南京)有限公司 | The machine failure diagnosis and prediction method and system cooperateed with based on edge calculations and cloud |
| CN110059601A (en) * | 2019-04-10 | 2019-07-26 | 西安交通大学 | A kind of multi-feature extraction and the intelligent failure diagnosis method merged |
| CN110394688A (en) * | 2019-09-02 | 2019-11-01 | 太原科技大学 | Conditions of machine tool monitoring method based on edge calculations |
| CN110646183A (en) * | 2019-09-04 | 2020-01-03 | 国网浙江省电力有限公司嘉兴供电公司 | High-voltage circuit breaker mechanical property quick detection device based on thing networking |
| CN111259927A (en) * | 2020-01-08 | 2020-06-09 | 西北工业大学 | Rocket engine fault diagnosis method based on neural network and evidence theory |
| CN111442926A (en) * | 2020-01-11 | 2020-07-24 | 哈尔滨理工大学 | Fault diagnosis method for rolling bearings of different models under variable load based on deep characteristic migration |
| CN111366816A (en) * | 2020-04-26 | 2020-07-03 | 华北电力大学 | Power grid fault diagnosis method based on machine learning |
| CN111651312A (en) * | 2020-07-28 | 2020-09-11 | 南京宏声科技有限公司 | Three-level bidirectional early warning method and system for equipment fault based on edge calculation |
Non-Patent Citations (2)
| Title |
|---|
| MA BIN ET AL.: "Application of D-S evidence theory in equipment fault diagnosis", 《2011 SEVENTH INTERNATIONAL CONFERENCE ON NATURAL COMPUTATION》 * |
| 张文龙 等: "云/边缘协同的轴承故障诊断方法", 《计算机集成制造系统》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113507491A (en) * | 2021-04-30 | 2021-10-15 | 华中农业大学 | Method and system for uploading all-chain information of clean egg production in real time |
| CN113507491B (en) * | 2021-04-30 | 2022-11-11 | 华中农业大学 | Method and system for uploading full-chain information of clean egg production in real time |
| CN113255584A (en) * | 2021-06-22 | 2021-08-13 | 德明通讯（上海）股份有限公司 | Fault diagnosis and monitoring system based on edge calculation |
| CN113255584B (en) * | 2021-06-22 | 2021-10-19 | 德明通讯（上海）股份有限公司 | Fault diagnosis and monitoring system based on edge calculation |
| CN113973029A (en) * | 2021-10-27 | 2022-01-25 | 广东电网有限责任公司江门供电局 | Low-voltage Internet of things edge computing intelligent gateway system and communication method |
| CN113973029B (en) * | 2021-10-27 | 2023-01-20 | 广东电网有限责任公司江门供电局 | Low-voltage Internet of things edge computing intelligent gateway system and communication method |
| CN114057053A (en) * | 2022-01-18 | 2022-02-18 | 杭州浅水数字技术有限公司 | Method for monitoring fatigue degree of component of special machine |
| CN114057053B (en) * | 2022-01-18 | 2022-04-26 | 杭州浅水数字技术有限公司 | Method for monitoring fatigue degree of component of special machine |
| CN115946132A (en) * | 2023-03-15 | 2023-04-11 | 江西丹巴赫机器人股份有限公司 | Intelligent dense warehouse and logistics robot system and fault monitoring method thereof |
| CN116467570A (en) * | 2023-06-16 | 2023-07-21 | 西南交通大学 | Heavy-duty locomotive coupler swing angle quantitative identification method based on data driving |
| CN116467570B (en) * | 2023-06-16 | 2023-09-05 | 西南交通大学 | Heavy-duty locomotive coupler swing angle quantitative identification method based on data driving |
Also Published As
| Publication number | Publication date |
|---|---|
| CN112216085B (en) | 2022-05-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN112216085B (en) | Equipment key load-bearing structural member health monitoring system based on edge calculation and online update sample intelligent identification | |
| CN111124840B (en) | Method and device for predicting alarm in business operation and maintenance and electronic equipment | |
| CN105550323B (en) | Load balance prediction method and prediction analyzer for distributed database | |
| CN110515732B (en) | Task allocation method based on deep learning inference of resource-constrained robot | |
| CN114721833B (en) | Intelligent cloud coordination method and device based on platform service type | |
| CN112101525A (en) | Method, device and system for designing neural network through NAS | |
| US20220164660A1 (en) | Method for determining a sensor configuration | |
| CN114815605A (en) | Automatic driving test case generation method and device, electronic equipment and storage medium | |
| CN108804473A (en) | The method, apparatus and Database Systems of data query | |
| CN112418482A (en) | Cloud computing energy consumption prediction method based on time series clustering | |
| CN114297036A (en) | Data processing method and device, electronic equipment and readable storage medium | |
| CN110543500A (en) | Power transmission and transformation equipment health assessment platform based on big data | |
| Ravichandran et al. | Efficient multilevel federated compressed reinforcement learning of smart homes using deep learning methods | |
| McNamee et al. | The case for adaptive deep neural networks in edge computing | |
| CN115051929A (en) | Network fault prediction method and device based on self-supervision target perception neural network | |
| CN112462759B (en) | Evaluation method, system and computer storage medium of rule control algorithm | |
| CN113541986B (en) | Fault prediction method and device for 5G slice and computing equipment | |
| Chandar et al. | Communication optimization in large scale federated learning using autoencoder compressed weight updates | |
| CN110784555A (en) | Intelligent monitoring and load scheduling method based on deep learning | |
| CN111209105A (en) | Capacity expansion processing method, capacity expansion processing device, capacity expansion processing equipment and readable storage medium | |
| CN115794586A (en) | Cloud server software aging prediction method, device, equipment and medium | |
| CN112906745B (en) | Integrity intelligent network training method based on edge cooperation | |
| Wang et al. | Edge-computing based soft sensors with local Finite Impulse Response models for vehicle wheel center loads estimation under multiple working conditions | |
| CN114444922A (en) | Hybrid traffic efficiency evaluation method under group intelligent control | |
| CN114139782A (en) | Traffic prediction method, device, equipment and computer readable storage medium |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |