THE STRUCTURAL DESIGN AND USE OF HIGHER FORMS OF CONTROL IN REHABILITATION DEVICES


Authors: Marián Veseliny 1;  Boris Jobbágy 2;  Marek Fodor 3;  Miloslav Feriančik 4
Authors‘ workplace: Mechanical Engineering Faculty, Technical University of Košice, Slovakia 1,2,3,4
Published in: Lékař a technika - Clinician and Technology No. 2, 2012, 42, 73-76
Category: Conference YBERC 2012

Overview

Although the area of artificial intelligence is still progressing, its use in different areas is still insufficient. This article discusses the use of higher forms of control in rehabilitation devices (specifically, the neural networks). Then the article deals with problems of structural design. This proposal is based on the requirements of the rehabilitation process, which must be guaranteed maximum patient safety. This proposed device uses various modern elements. The device will be powered by unconventional actuator, represented by pneumatic artificial muscles.

Keywords:
Artificial intelligence, neural networks, JavaNNS, pneumatic artificial muscles, rehabilitation device, construction, exoskeleton

Introduction

The article discusses the use of artificial intelligence in rehabilitation devices. Modern rehabilitation device is designed to replace the work of rehabilitation staff. The process of rehabilitation is increasingly being provided by rehabilitation staff. With the increase in human population is continuously increasing the number of patients. One solution to this situation is a modern rehabilitation device to facilitate the work of rehabilitation staff, or replace them completely. Rehabilitation itself is time-consuming process and personnel. It affects a number of factors such as patient health status, type of damage etc.

This article describes the possibilities of artificial intelligence in modern rehabilitation devices. It is mainly the implementation of neural networks to such device. The role of the neural network is to replace a part the decision in the rehabilitation process. Using a variety of inputs is designed to intelligently load change or rehabilitation program.

Simulations of neural networks were created in JavaNNS. Modern rehabilitation device created by combining a variety of advanced features, such as artificial intelligence and advanced unconventional propulsions. One of these actuators are also pneumatic artificial muscles. Their properties are broadly similar to human muscles. They are therefore suitable for use in the rehabilitation area. Soft and smooth movement needed for the rehabilitation can be done just by pneumatic artificial muscles. This will eliminate the need for constant presence of a rehabilitation staff. Furthermore, the article deals with the design variant design rehabilitation device for the rehabilitation of upper limb. It also deals two proposed design options rehabilitation device.

Rehabilitation devices in present

In spite of major advances in the section of automation is currently in rehabilitation performed principally with the help of rehabilitation operator. This is of time consuming and unrealistic on given capacity. One solution of this situation is research automated rehabilitation devices, which replace work of rehabilitation worker. We know two method of rehabilitation, they are active or passive. With active rehabilitation the patient participate on implementation of exist movement, for example works against movement robot or copying their movement. With passive rehabilitation only robot perform movement, the patient does not perform any movement. Their nervous system "teach in" motion.

Most rehabilitation devices in present focuses on specifically muscle parts and want complex equipment, which could be used to rehabilitate more parts of the body. Essentially we can divide rehabilitation devices onto some groups:

Under rehabilitation body of part, on rehabilitation device practice:

  • upper limbs
  • lower limbs
  • thorax

Under load on rehabilitation device:

  • with zero initial load (suitable for patient after operation)
  • with nonzero initial load

Rehabilitation device for rehabilitation upper limbs

Robotic device for rehabilitation of upper limb can be divided into:

  • JOYSTICK
  • EXOSKELETON

In the first case the patient is secured to the arm structure similar to a joystick. Joystick / lever arm to lead positions in accordance with training plan. In the latter case is a construction attached to the patient. The upper extremity is placed in the exoskeleton. The rehabilitation device may be a stand that serves as a carrier of the structure.

The structural design of rehabilitation device

In order to propose the construction of rehabilitation device, we must first know the requirements of the rehabilitation process. In our case, the rehabilitation of upper limb, which will be powered by artificial muscles. Since we want to rehabilitate the upper limbs, it is appropriate to the patient during rehabilitation was sitting. The draft design options are considered and appropriate design seats. The following steps are described in two variants. Before proposing the first variant, we know design requirements were as follows:

  • design must be Easy to install
  • for the rehabilitation is necessary to reach 4 degrees of freedom (arm joint 3 ° and 1 ° elbow)
  • the possibility of rehabilitation of both upper limbs •configurability device applicable to different populations.

This structure (shown in Figure No. 1) consists of standard aluminum profiles, which are in terms of low weight and simple application suitable. Aluminium profiles are connected with each part of the adjustment, which allows setting of different sizes of the upper limb. Since this is a rehabilitation device based on artificial muscles, it is necessary to solve the structural design of storage (PAM) muscles. In the design of variant 1 are placed pneumatic artificial muscles at the back of seats and rotation of the joints will be performed by a steel cable attached to the roller. [1]

Fig.1: The structural design variants No.1
Fig.1: The structural design variants No.1

Variant No. 2 is also designed on the principle of exoskeleton. The device has 4 degrees of freedom of movement. Each joint is powered by a pair of pneumatic muscles that are involved in antagonistic and are connected at joints with cable transfers. Muscles for each joint are placed directly in the design of exoskeleton. Using screws, the exoskeleton can be adapted for different size limbs.

Figure No. 2 shows a patient who has a construction rehabilitation devices connected to the right hand. The device is suspended on adjustable stand. 


Fig. 2: The structural design variants No.2
Fig. 2: The structural design variants No.2

Modern rehabilitation devices

Still missing a device that can respond flexibly to change the various conditions. It therefore seems to be a combination of rehabilitation devices with artificial intelligence, like a good solution. The combination of unconventional actuators and artificial intelligence can build the future of modern rehabilitation device that can respond flexibly to a variety of new situations.

Figure No. 3 shows a modern rehabilitation device that uses elements of artificial intelligence and unconventional actuators. Description of device:

CS – Control system

NN – Neural network

Electro-pneumatics circuits 

Fig. 3: Modern rehabilitation device, using artificial intelligence and unconventional drives
Fig. 3: Modern rehabilitation device, using artificial intelligence and unconventional drives

The device works as follows: Information is gathered by various sensors. Then they are sent to the control system and neural network. NN gives rise to a control system that sends a control instruction for electropneumatic circuits that perform the actual movement of a rehabilitation device. Movement a rehabilitation device is represented by pneumatic artificial muscles. Rehabilitation device must of course have various security features. An example of a security system is described in the publication: “Real-time bezpečnostný systém v automatizovaných rehabilitačných zariadeniach” [2].

Control system

An example of a control system without the neural network is shown in Fig. (Fig. 4) 

Fig. 4: Blok diagram of control part
Fig. 4: Blok diagram of control part

The activity of the proposed control system is described by other investigators of the project. Part of the proposed control system is described in: “Design of the control system for rehabilitation device of upper arm” [3].

The most important part of the inputs to the control system mainly consists of inputs from sensors. Sensors placed directly onto the structure, or outside, provide important information needed for control. These sensors include:

  • Pressure sensor
  • Temperature sensor
  • Sensor of dangerous acceleration (Accelerometer)
  • Speed sensor (Incremental)
  • Sensor of dangerous rotation (Gyroscope)
  • and other

Neuron networks in a rehabilitation devices

One possible application of artificial intelligence in a rehabilitation device is the use of neural networks (NN) for the control of a rehabilitation device. It is a part of the control system. NN with input by the sensors provides information control system that performs the actual action and controls rehabilitation device. The NN control system is able to evaluate whether a change is needed load. Modern rehabilitation device is available from the beginning of rehabilitation and is also suitable for patients immediately after surgery. It is used mainly feature NN learn during the process. While rehabilitation is able to detect improvement of physical properties and react by increasing patient load.

Inputs and outputs

When designing a NN learning is the most important definition of inputs and outputs related to them. Neural network decides on the basis of these parameters: speed, load, direction, angle and others. Using these inputs neural network generates outputs.

Simulation of neural network in JavaNNS

The JavaNNS (JavaNeuralNetwork Simulator) is a program based on its predecessor SNNS (StuttgartNeuralNetwork Simulator). This is the NN simulator. In JavaNNS is improved graphical interface and is much simpler than its predecessor. However, some parts were omitted.

The actual simulation is performed in the following steps:

  • Preparation learning and test data
  • The design of NN
  • NN learning
  • Testing the NN

Preparation of learning data is very difficult because it determines the entire functioning of the neural network. Design of the neural network is thanks to graphic interface simpler and more transparent. The proposed NN is shown in (Fig. 5): 

Fig. 5: The proposed NN with three hidden layer
Fig. 5: The proposed NN with three hidden layer

The learning we can use different kinds of learning functions and the process of learning affects many adjustable parameters. Learning is assessed according to NN learning errors, which is shown in Error graph (Fig. 6).

Fig. 6: Error graph with more learning function
Fig. 6: Error graph with more learning function

The picture (Fig. 10) shows the error graph, which is the result of NN learning using various learning functions. Thus, the proposed NN is able to learn with an acceptable error learning. After learn NN comes the testing process. Evaluation of the NN testing performed on a comparison of actual and expected results. [4]

Conclusion

The combination of artificial muscles and elements of artificial intelligence seems like a promising development in modern rehabilitation device. This proposed rehabilitation device operates largely autonomously. These options meet the design requirements necessary for the rehabilitation process, which was previously entered. Both variants were simulated in the CAD programs, and they allow mobility shoulder and elbow. These devices are suitable for active and passive rehabilitation. Mastering the nonlinearity of pneumatic artificial muscles require advanced management systems. Using artificial intelligence, therefore, appears to be very appropriate. It is not however the only neural networks, but the application can certainly find other elements of artificial intelligence, such as neuro-fuzzy systems,genetic algorithms and fuzzy relational system. Simulation of the proposed neural network has demonstrated the possibility of using artificial intelligence in a rehabilitation device is a realistic perspective.

Acknowledgement

VEGA 1/1162/11 Theoretical principles, methods and instruments of diagnostics a rehabilitation of senior mobility, coordinator: prof. Ing. Dušan Šimšík, PhD.

The research work is supported by the Project of the Structural Funds of the EU, Operational Program Research and Development, Measure 2.2 Transfer of knowledge and technology from research and development into practice: Title of the project: Research and development of the intelligent nonconventional actuators based on artificial muscles ITMS code: 26220220103. 


We are support research activities in Slovakia / Project is cofounded from sources of ES.

Ing. Marián Veseliny

Department of Automation, Control and Human

Machine Interaction

Mechanical Engineering Faculty Technical University of Košice

Letná 9, 042 00 Košice

E-mail: marian.veseliny@tuke.sk

Phone: +421 55 602


Sources

[1] PITEĽ, J., BALARA, M., BORŽÍKOVÁ, J. :Control of the actuator with pneumatic artificial muscles in antagonistic connection. Sborník vědeckých prací Vysoké školy báňské - Technické univerzity Ostrava. Vol. 53, no. 2 (2007), p. 101-106, ISSN 1210-0471

[2] RIGASOVÁ, Eva - ŽIDEK, Kamil, Real-time bezpečnostný systém v automatizovaných rehabilitačných zariadeniach, elektronický optický disk (CD-ROM). In: Automatizácia a riadenie v teórii a praxi 2012 : ARTEP 2012 : workshop odborníkov z univerzít, vysokých škôl a praxe v oblasti automatizácie a riadenia : 22. - 24. február 2012, Stará Lesná, SR. Košice : TU, 2012 S. 54-1-54-7. ISBN 978-80-553-0835-7

[3] ŽUPA, Tomáš - ŽIDEK, Kamil - LÍŠKA, Ondrej, Design of the control system for rehabilitation device of upper arm, Engineering Mechanics 2011 : 17th international conference : May 9-12, 2011, Svratka, Czech Republic. - Žďár nad Sázavou : ŽĎAS, 2011 P. 703-706. - ISBN 978-80-87012-33-8

[4] FISCHER, Igor, HENNECKE, Fabian, BANNES, Christian, ZELL, Andreas: Java Neural Network Simulator - User Manual, Version 1.1, Dostupné na internete: http://www.ra.cs.unituebingen. de/software/JavaNNS/manual/JavaNNS-manual.html

Labels
Biomedicine

Article was published in

The Clinician and Technology Journal

Issue 2

2012 Issue 2

Most read in this issue
Login
Forgotten password

Don‘t have an account?  Create new account

Forgotten password

Enter the email address that you registered with. We will send you instructions on how to set a new password.

Login

Don‘t have an account?  Create new account