Dynam ic GRNNmodel of hydraulic system fault detection based on the internet of things technology

Jie SUN，Qing-he XU*，Ye SUN，Zhi-jun LIU，Li-feng ZHANG，Yu SUN，Rui-xin ZHANG

（1 College of Electrical Engineering，North China University of Science and Technology，Tangshan 063210，China）

（2 BeijingTelecom Planning and Design Institute Smart City Design Institute，Beijing 100000，China）

（3 Tianjing Tiangang United Special Steel Co.，Ltd.，Tianjing 301500，China）

（4 Tangshan Ruifeng Iron&Steel（Group）Co.，Ltd.，Tangshan 063000，China）

（5 Ke Mei Rui（Tangshan）Environmental Protection Technology Co.，Ltd.，Tangshan 063000，China）

Abstract：The operation safety and conditionmonitoring of hydraulic equipment is an important content in production because of its unique characteristics，hydraulic system iswidely used in various fields.It is difficult to locate the fault source because all the components of the hydraulic system work in the closed oil circuit.In order to solve this problem，this essay puts forward a kind of fusion of Internet of things technology dynamic fault diagnosis method of the GRNNmodel，which is based on intelligent control of the Internet of things and using of wireless sensor network technology in distributed hydraulic equipmentof each parameter in real-timemeasurementand control，remote data sharing，the acquisition of fault signal input GRNN model fault observer，detection threshold，through experimental simulation accurate diagnosis of system failure.In response to this problem，a fault diagnosis method of the dynamic GRNNmodel incorporating the Internetof Things technology is proposed.Based on the intelligent control of the Internet of Things，the wireless sensor network technology is used tomeasure and control the data parameters of the distributed hydraulic equipment in real time and remotely.Thismethod inputs the collected fault signal into the GRNNmodel fault observer，calculates the detection threshold，and accurately diagnoses the system fault through experimental simulation.The experiment showsthat thismethod can be effectively applied in the hydraulic system of process production to ensure the normal operation of the system，reduce the failure rate of equipment，and improve the production efficiency.

Key words：The hydraulic system，The internet of things，Wireless sensor network，Dynamic GRNN model，Fault diagnosis

1 Introduce

The hydraulic system is an important part ofmodern mechanical equipment，whose the stability and reliability of equipment operation are closely related to the hydraulic system.In the actual production process，it iswidely used in various fields because of the unique characteristics of the hydraulic system：extensive process adaptability，excellent control performance and low cost.

To begin with，ensuring the normal operation of the hydraulic system and reducing the failure rate of the hydraulic equipment are the core requirements in the actual production process.How to determine the fault and achieve fault prediction is the goal of research［1］.Due to the nature of the hydraulic system，it is difficult to locate the fault source，which brings some limitations to the ordinary work.The introduce of hydraulic system fault detection research can not only improve the efficiency in the actual production，but also reduce the hydraulic equipment failure rate and the production safety is of great practical significance.

Recently，there are in-depth research on hydraulic fault detection［2］.Part of the method is using support vectormachine（SVM）to realize fault detection.However，thismethod has an obvious limitation that it can’t detect the small fault through the hydraulic system of the initial signal.So the signal characteristic classification［3］is to realize the fault detection of the hydraulic system.

With the high speed of development of smartmanufacturing and Internet technology，remote monitoring of devices has become a development trend［4］.Traditional fault detection and maintenance requiremanual operation on site，resulting in low efficiency and other problems.The use of Internet of Things technology in hydraulic fault diagnosis can realize the remote transmission of hydraulic system fault information and remotemaintenance，andsolve fault problems.It not only improveswork efficiency，but also saves travel service fees.

For the problems above，a generalized regression neural network（GRNN）model is proposed for fault detection of hydraulic system［5］，which is based on the Internet of Things technology.Themain idea is to solve the remote maintenance and realize the overall management of the equipment in the actual production process.The adaptive threshold mechanism introduced by the dynamic GRNN model can effectively improve the fault detection efficiency［6］and realize the detection of small amplitude abrupt faults and early slowvarying faults of hydraulic equipment.

2 Dynam ic GRNN m odel based on In ternet of things technology research overall design hyd rau lic system fau lt detection

2.1 Summary of the internet of things

With the continuous development of Internet of things technology，the connection of things and data sharing have become a reality.In production and life，remote control has become the trend of the production and comprehensive management reform of various enterprises.For example，intelligent home，unmanned vehicle and remote control of equipment have been realized［7-8］.

In sewage treatment，there is still a gap in realizing the control of each equipment in the hydraulic system，real-timemonitoring of each data parameter，and realizing the connection of things and comprehensive control.In order to ensure the normal operation of the hydraulic system，it is themost important thing in sewage treatment process.The introduction of Internet of Things technology is to better realize the real-time status and fault signal acquisition of hydraulic equipment，predict abnormal status of equipment in advance，reduce equipment failure rate，ensure production development，and improve production efficiency.

2.2Wireless Sensor Network

In intelligent production，Wireless Sensor Networks（WSN），as a core technology［9-11］，is widely used in industrial production，such as the operation of sensing environmental variables and relevant data processing.It has the advantages of low cost，low power consumption and the realization of distributed device AD hoc network.Distribution on the hydraulic equipment，a large number of sensors，obtained the running status of equipment parameters，and data communication technology in wireless sensor network data transmission，signal the bidirectional interaction，can realize equipment failure in time，provide safe and reliable service for the hydraulic system，can realize something connected between the devices，realize the management as a whole.The structure diagram of wireless sensor network for distributed hydraulic equipment is shown in Fig.1.

Fig.1 Structure diagram of w ire Iess sensor network fo r d istribu ted hyd rau Iic equipm en t

2.3 Overall design scheme of the system

In terms of technology architecture，the Internet of things ismainly divided into parameter information acquisition layer；data signal transmission layer and comprehensive analysis and diagnosis layer［12-13］.Firstly，the parameter information acquisition layer uses the wireless sensor network：the sensor device is installed in the hydraulic system to collect data.Secondly，data signal transmission layer：realize remote monitoring of the operation status of the hydraulic system by transferring the important data of distributed equipment，such as pressure and flow，to the monitoring interface［14］.Finally，comprehensive analysis and diagnosis layer：through the dynamic GRNNmodel，the input dynamic signals and data can be quickly diagnosed and analyzed to obtain the target fault，and the correspondingmaintenance strategy can be provided.The overall scheme design of the system is shown in Fig.2.

Fig.2 Overa IIscheme design d raw ing of the system

Initially，the wireless sensor network composed of a large number ofwireless sensors is used to collect and monitor the data of the hydraulic equipmentat the bottom，so that the system canmonitor the operating state and fault signals of the hydraulic system and equipment.

Apart from that，itwill collect data signal transmission to the cloud platform of information integrated processing and using dynamic GRNNmodel analysis of fault signal diagnosis through the Internet communication technology.The hydraulic equipment failure condition［15-19］makes the corresponding solution to solve the failure problems of hydraulic system，and to realize remotemonitoring and something connected between the devices.

2.4 Data signal acquisition research and design of hydraulic system

In a hydraulic system，we need to collect the two critical data parameters：flow and pressure.The fault diagnosis of hydraulic system is carried out by judging whether these two parameters are normal.

As can be found from Fig.3，1YA～4YA is the mainmeasuring pointelementof the hydraulic system，which are the directional control valve and the electromagnet coil.The flow sensor is placed at the①monitoring point to detect the oil tank flow of the hydraulic system.Pressure sensors and flow sensors are set at the positions shown in the②-⑧ respectively.Pressure sensors are placed at themonitoring points②to detect the pressure generated by the amount of oil returned.Themonitoring point③measures the flow of the hydraulic pump into the upper chamber of the hydraulic cylinder and the pressure holding；.Themonitoring points④ and⑤detect the pressure and flow rate of the reversing valve 3，so as to monitor the working state of the directional control valve 3.The following part is monitoring point⑥monitoring the working status of the hydraulic pump in the hydraulic system.The working state of electric directional control valve 2 and hydraulic cylinder 1 is tested by pressure and flow sensors atmonitoring points⑦and⑧.The fault diagnosis of hydraulic system is realized by combining the collected signals and parameters with GRNNmode.

Fig.3 Schem atic d iag ram of measuring points distribution of m ain com ponents of the hydrau Iicsystem

2.5 Experimental study on data transmission in hydraulic system

The data signal transmission layer is the core of the whole system，ensuring the real-time，continuity and security of the collected signal parameter transmission，and preventing data loss and interruption of transmission.The components of the hydraulic system work in a closed environment.The signal parameters collected by the sensor are transmitted in time，which plays an important role in fault diagnosis.A short sampling period is set to solve the differences in data acquisition and processing，so as to make the signal parameter transmission timely.The data acquisition process is shown in the Fig.4.

Fig.4 FIow chart of data co IIection

2.6 Data transmission performance test

In the actual production process of the hydraulic system，therewill be some fault problems，resulting in the hidden danger of the equipment being disconnected and offline in the process of data transmission.To prevent interruption of data transfer，we should make sure that the data transfer can be sent“at least once”.By introducing the cloud platform，the server can realize faster response to process data than the local server.Through the data transmission system，the same number of packets were transmitted to the cloud platform and the local server respectively，and through the comparison and analysis of the time consumed，it was concluded that the time spent by the local server to receive the packets was longer than the time taken by the cloud platform server to receive the packets，with an average delay difference of about 0.6 s.As shown in the Fig.5，the cloud platform server is superior to the local server in data storage and transmission.

Fig.5 Com parison diagram of packet transm ission perfo rm ance

3 Dynam ic GRNN neuralm odel theory

3.1 Generalized regression neural network（GRNN）

Generalized regression neural network is a radial basis function network based on mathematical statistics，and its theoretical basis is nonlinear regression analysis［20］.GRNN algorithm has great advantages in learning speed and nonlinear mapping abilityand is suitable for dealing with nonlinear problems.The algorithm has the advantages of training speed block，high approximation ability and high simulation accuracy.

Assuming the joint probability densityf（x，y）of the random vectorxand the random vectory，the regression ofywith respect toxcan be expressed as：Where，the estimation of density functionf（x，y）can be obtained by using Parzen’s non-parametric estimator on the training data，and its non-parametric estimation is：

Among them：

Wherexirepresents the observation vector ofx；yiis the observed value ofy；σis the smoothing coefficient；nis the number of samples；mis the dimension of the random vectorx.

As can be seen from equation（6），（x）is the weighted average of all the observed values ofyi，and the weight factor of each observed valueyiis the exponent of the Euclid distance square between the corresponding samplexiandx.

3.2 Dynamic GRNNmodel structure

A generalized regression neural network is constructed according to equation（6），and the structure is shown in Fig.6.Thewhole structure consists of four layers of neurons：Input layer，Pattern layer，Summation layer and Output layer.

Fig.6 GRNN structure diagram

4 Hyd rau lic system fau lt detection research

4.1 Build a fault observer for GRNNmodel

In the environment of hydraulic system，the sensor network composed of various sensors is used to acquire the important signals in the hydraulic equipment.Through the GRNN observermodel training，the residual of the hydraulic system is obtained，and the dynamic GRNN model is constructed.Fig.7 shows the structure diagram of dynamic GRNN model.

Fig.7 Dynam ic GRNN structure diagram

Through dynamic GRNN structure diagram，we can clearly get the network model expressions of input and output.

The dynamic GRNN model belongs to the global recursivemodel，which often involves the prediction of themodel，that is，the model prediction output value at the previousmoment becomes themodel input value at the nextmoment.General prediction can be divided into single-step prediction and multi-step prediction.For dynamic GRNNmodel，themain predictionmethod ismulti-step prediction.

4.2 Based on the dynamic GRNN model of hydraulic system fault detection research

Combined with the experience of fault detection of hydraulic equipment in sewage treatment，it can be seen that in sewage treatment in different environments，random variables and random interference have mutual influence，which will lead to different kinds of fault states.

Through a normal sample input signal，we will input dynamic GRNN model and get a model output.Through the collected in the actual production of signal input，themodelwill also getan actual output values and the normal model output value，compared with the actual output is a residual value，and the residual value controlwithin the threshold we set.If hydraulic equipmentmalfunction，leads to a failure state signal input，output produced by themodelwith residual error between actual output can produce larger with the previously set threshold value deviation，so will detect the equipment failure.

The fault detection of hydraulic equipment is described in the form ofmathematicalmodel.The system residual sequence is as follows：

With^ysaid dynamic GRNNmodels to predict the output，ysaid sewage treatment in the production of the actual output.

Formula（10）is used to calculate the square sum of the residual sequence andJ（k+1），and N represents the window length.The following are two cases of the size of theNvalue：

By setting the fault detection threshold of hydraulic system for sewage treatment asε0，ε0is usually set at different thresholds in different environments.Normally，2 to 3 times of the sum of squares of the normal sample residuals ofJ0，that isε0=2J0，we can obtain the discriminant of equipment fault detection：

In the environment of sewage treatment，when the fault problem of hydraulic system is diagnosed by the dynamic GRNN model，the severity of the fault of hydraulic system can be judged by the size ofJfvalue.

5 Experim ental resu lts and analysis

The simulation platform of GRNN fault observerwas builtby using MATLAB to realize the state fault detection of hydraulic equipment in the hydraulic system.Experiments that the experiment signal input by flow sensor pressure sensors for data acquisition.In the hydraulic system，30 groups of samples were collected for the experimental signals to be collected，such as outlet flow，outlet pressure and pressure of the hydraulic cylinder chamber.The dynamic GRNN model of hydraulic system was constructed by using the collected signal samples，and six target types of hydraulic system were set.Through the dynamic GRNN model，we can know the detection threshold under normal state.Ten groups of samples are taken as test samples from the collected samples，and then the diagnosis results are observed through MATLAB simulation.As shown in Fig.8，we can clearly detectwhere the fault point is.When the same target fault threshold is less than the corresponding detection threshold，the point is regard asfault point.

It isworth noting that in Fig.8（g），and Fig.8（f），there are two types of target faults that are less than the detection threshold.We only need to compare the threshold values of these two types，and determine which type of threshold is smaller，and then determine the point as the fault point.By testing all the collected samples，the surfacemethod can be effectively applied to the fault detection of hydraulic system.

In the real process production，the hydraulic system parameter collection and monitoring often can not achieve real-time synchronization，so it is difficult to predict in advance.Along with the development of the Internet of things technology，wireless sensor network technology matures，connected the equipment to achieve something in hydraulic system，parameter of the hydraulic system to realize real-time monitoring and gathering，through dynamic GRNN model analysis，observation in advance that the hydraulic system of abnormal state，fast realization of hydraulic equipment fault diagnosis，accordingly find out the cause of the problem lies.

Fig.8 Target fau It detection diagram

6 Conclusions

In view of the actual process during the production of a hydraulic system failure problems caused by environment，we introduced a dynamic GRNN model of hydraulic system fault detectionmethod.It involvesthe signs of the Internet of things technology to provide comprehensive information fusion and fault signal，and usesthe simulation software to realize fault diagnosis，reduce diagnostic uncertainty，and improve the diagnostic accuracy and reliability.

Based on the intelligent control of the Internet of things，the integrated fault detection system of the Internet of things can solve the problem of cooperation and security between hydraulic devices.In addition，the wireless sensor network can be used to enhance the security of data signal collection，greatly reduce the operating cost of the Internet of things，and realize fault diagnosis and remotemaintenance of the hydraulic system.In actual production，maintenance engineers can reduce the on-site maintenance method，which canguarantee the use capacity of products，improve the production efficiency of water plant and maintain the working capacity of equipment.What is more，itmeets the future development needs of enterprises brings a new way of development to other industries.