Search results

The purpose of this paper is to present a multi‐expert system that can provide designers with suggestions for improvement. The multi‐expert system can analyse a design and provide designers with ideas for changes to designs at an early stage in order to improve assembly later in the manufacturing process.

The whole system consists of four expert systems: computer‐aided design (CAD) expert, automated assembly expert, manual assembly expert and design analysis expert. The design analysis expert includes a sub‐system to collate the information from the assembly experts and to provide costs and advice.

The paper finds that the approach and the systems can reduce manufacturing costs and lead times.

A knowledge‐based reckoning approach to design‐for‐assembly automation is used. The approach and systems can reduce manufacturing costs and lead times. The system can estimate assembly time and cost for manual or automatic assembly and select suitable assembly techniques.

The system can estimate assembly time and cost for manual or automatic assembly and select a suitable assembly technique.

The new system models assembly, product and process design using a natural approach for capturing intelligence. The new approach categorised automated assembly and manual assembly into separate individual experts. Intelligence and knowledge from each is captured and embedded within the individual expert that represented the process. This approach enabled greater flexibility and made the sub‐systems easier to modify, upgrade, extend and reuse.

The purpose of this paper is to investigate how to make powered‐wheelchair driving easier using simple expert systems to interpret joystick and ultrasonic sensor data. The expert systems interpret shaky joystick movement and identify potentially hazardous situations and then recommend safe courses of action.

The way that a human user interacts with a powered‐wheelchair is investigated. Some simple expert systems are presented that interpret hand tremor and provide joystick position signals for an ultrasonic sensor system. Results are presented from a series of timed tasks completed by users using a joystick to control a powered‐wheelchair. Effect on the efficiency of driving a powered‐wheelchair is measured using the times to drive through progressively more complicated courses. Drivers completed tests both with and without sensors and the most recently published systems are used to compare results.

The new expert systems consistently out‐performed the most recently published systems. A minor secondary result was that in simple environments, wheelchair drivers tended to perform better without any sensor system to assist them but in more complicated environments then they performed better with the sensor systems.

The time taken for a powered‐wheelchair to move from one place to another partly depends on how a human user interacts with the powered‐wheelchair. Wheelchair driving relies heavily on visual feedback and the experience of the drivers. Although attempts were made to remove variation in skill levels by using sets of data associated with each driver and then using paired statistical tests on those sets, some variation must still be present.

The paper presents new systems that could allow more people to use powered‐wheelchairs and also suggests that the amount of sensor support should be varied depending on circumstances.

The new systems described in the paper consistently performed driving tasks more quickly than the most recently published systems.

The paper aims to propose a system that uses a combination of techniques to suggest weld requirements for ships parts. These suggestions are evaluated, decisions are made and then weld parameters are sent to a program generator.

A pattern recognition system recognizes shipbuilding parts using shape contour information. Fourier‐descriptors provide information and neural networks make decisions about shapes.

The system has distinguished between various parts and programs have been generated so that the methods have proved to be valid approaches.

The new system used a rudimentary curvature metric that measured Euclidean distance between two points in a window but the improved accuracy and ease of implementation can benefit other applications concerning curve approximation, node tracing, and image processing, but especially in identifying images of manufactured parts with distinct corners.

A new proposed system has been presented that uses image processing techniques in combination with a computer‐aided design model to provide information to a multi‐intelligent decision module. This module will use different criteria to determine a best weld path. Once the weld path has been determined then the program generator and post‐processor can be used to send a compatible program to the robot controller. The progress so far is described.

This paper aims to describe real time improvements to the performance and trajectories of robots for which paths had already been planned by some means, automatic or otherwise. The techniques are applied to industrial robots during the gross motions associated with pick and place tasks. Simple rules for path improvement are described.

The dynamics of the manipulator in closed form Lagrange equations are used to represent the dynamics by a set of second‐order coupled non‐linear differential equations. The form of these equations is exploited in an attempt to establish some simple rules. Sub‐optimal paths are improved by considering simple rules developed from the model of the machinery dynamics. By considering the physical limitations of the manipulator, performance was improved by refining pre‐calculated paths. Experiments were performed with a prototype robot and an old Puma 560 robot in a laboratory environment. Once the method had been tested successfully then experiments were conducted with a Kuka KR125 Robot at Ford Motor Company. The measured quantities for all the robots were drive currents to the motors (which represented the torques) and the joint angular positions.

The method of path refinement presented in this paper uses a simplified model of the robot dynamics to successfully improve the gross motions associated with a pick and place task. The advantage of using the input‐output form described was that intermediate non‐linearities (such as gear friction) and the motor characteristics were directly incorporated into the model.

Even though many of the theoretical problems in manipulator dynamics have been solved, the question of how to best apply the theories to industrial manipulators is still being debated. In the work presented in this paper, information on system dynamics is used to produce simple rules for “path improvement”.

Most fast algorithms are for mobile robots and algorithms are scarcer for manipulators with revolute joints (the most popular type of industrial robot). This work presents real time methods that allow the robot to continue working while new global paths are automatically planned and improved as necessary.

Motion planning for manipulators with many degrees of freedom is a complex task and research in this area has been mostly restricted to static environments, offline simulation or virtual environments. This research is applied in real time to industrial robots with revolute joints.

The purpose of this paper is to investigate how to make tele‐operated tasks easier using an expert system to interpret joystick and sensor data.

Current tele‐operated systems tend to rely heavily on visual feedback and experienced operators. Simple expert systems improve the interaction between an operator and a tele‐operated mobile‐robot using ultrasonic sensors. Systems identify potentially hazardous situations and recommend safe courses of action. Because pairs of tests and results took place, it was possible to use a paired‐samples statistical test.

Results are presented from a series of timed tasks completed by tele‐operators using a joystick to control a mobile‐robot via an umbilical cable. Tele‐operators completed tests both with and without sensors and with and without the new expert system and using a recently published system to compare results. The t‐test was used to compare the means of the samples in the results.

Time taken to complete a tele‐operated task with a mobile‐robot partly depends on how a human operator interacts with the mobile‐robot. Information about the environment was restricted and more effective control of the mobile‐robot could have been achieved if more information about the environment had been available, especially in tight spaces. With more information available for analysis, the central processor could have had tighter control of robot movements. Simple joysticks were used for the test and they could be replaced by more complicated haptic devices. Finally, each individual set of tests was not necessarily statistically significant so that caution was required before generalising the results.

The new systems described here consistently performed tasks more quickly than simple tele‐operated systems with or without sensors to assist. The paper also suggests that the amount of sensor support should be varied depending on circumstances. The paired samples test was used because people (tele‐operators) were inherently variable. Pairing removed much of that random variability. When results were analysed using a paired‐samples statistical test then results were statistically significant. The new systems described in this paper were significantly better at p<0.05 (95 per cent probability that this result would not occur by chance alone).

The paper shows that the new system performed every test faster on average than a recently published system used to compare the results.

The purpose of this paper is to present powered‐wheelchair transducers and systems that provide more control, reduced veer on slopes, and improved energy conservation, while reducing effort. They are especially significant for people with movement disorders who lack sufficient hand‐grasp and release ability or sufficient targeting skill to use joysticks.

Laboratory test rigs are created to test proportional switches and teach potential users. Then, trials are conducted with a rolling road and in real situations. Caster angle‐measurement is selected to provide feedback to minimize drift away from a chosen course and an electronic solution was created to match driver control to caster‐steering‐position. A case study is described as an example.

Results and advantages are presented from changing from using a set of digital‐switches to a set of new variable‐switches and then adding a sensor system to prevent veer on slopes. Systems have been tested for nearly two years and shown to assist powered‐wheelchair‐users with poor targeting skills.

The research used wheelchairs with caster‐wheels but the systems could easily be used on other wheelchairs.

Simple input‐devices are presented that isolate gross motor function and are tolerant to involuntary movements (proportional‐switches). A sensor system is presented that assists users in steering across sloping or uneven ground.

Proportional‐switches and sensors are shown to reduce veer and provide more control over turn and forward speed and turn radius while reducing frustration and improving energy conservation. The simple and affordable systems could be created and attached to many standard powered‐wheelchairs in many organisations.

The purpose of this paper is to investigate the effect on completion of mobile‐robot tasks depending on how a human tele‐operator interacts with a sensor system and a mobile‐robot.

Interaction is investigated using two mobile‐robot systems, three different ways of interacting with the robots and several different environments of increasing complexity. In each case, the operation is investigated with and without sensor systems to assist an operator to move a robot through narrower and narrower gaps and in completing progressively more complicated driving tasks. Tele‐operators used a joystick and either watched the robot while operating it, or sat at a computer and viewed scenes remotely on a screen. Cameras are either mounted on the robot to view the space ahead of the robot or mounted remotely so that they viewed both the environment and robot. Every test is compared with sensor systems engaged and with them disconnected.

A main conclusion is that human tele‐operators perform better without the assistance of sensor systems in simple environments and in those cases it may be better to switch‐off the sensor systems or reduce their effect. In addition, tele‐operators sometimes performed better with a camera mounted on the robot compared with pre‐mounted cameras observing the environment (but that depended on tasks being performed).

Tele‐operators completed tests both with and without sensors. One robot system used an umbilical cable and one used a radio link.

The paper quantifies the difference between tele‐operation control and sensor‐assisted control when a robot passes through narrow passages. This could be an useful information when system designers decide if a system should be tele‐operated, automatic or sensor‐assisted. The paper suggests that in simple environments then the amount of sensor support should be small but in more complicated environments then more sensor support needs to be provided.

The paper investigates the effect of completing mobile‐robot tasks depending on whether a human tele‐operator uses a sensor system or not and how they interact with the sensor system and the mobile‐robot. The paper presents the results from investigations using two mobile‐robot systems, three different ways of interacting with the robots and several different environments of increasing complexity. The change in the ability of a human operator to complete progressively more complicated driving tasks with and without a sensor system is presented and the human tele‐operators performed better without the assistance of sensor systems in simple environments.

A new type of high‐level robot command library is presented, which can be viewed as a marriage between simulation and control. The library commands contain simulations of the physical abilities of the robots as well as having the ability to control the physical machinery. The control of the machinery is performed by translating parameter information and then mapping the library commands to the robot controller commands. To demonstrate the use of the libraries, two robot programming languages have been analysed and new robot command libraries created for two types of machine. The robots selected were a Fanuc A600 and a Unimation PUMA robot. The paper also presents criteria that were used for assessing programming languages for use in programming and controlling robots. The paper shows how simulation can be incorporated into a high‐level robot command library (or object library) and how the command library can be used for the programming of industrial robots. The work has demonstated the advantages of including simulation within robot command libraries. The purpose of the research has not been to define another new robot command library, and the techniques presented here can be applied to other robot languages and high level robot command libraries.

A novel sensor system is described that can replace the need to use a touch screen, mouse or keyboard for use with either a thin film transistor (TFT) or cathode ray tube (CRT) screen.

A tri‐colour photodiode in a colour sensor pen (CSP) is used as an input to depict where a user is pointing by detecting a dynamic kaleidoscope of colours flashed onto the screen. The most easily detected colours are placed around the predicted position of the sensor. The output from the sensor is filtered and transmitted to a computer where colour is translated into a corresponding co‐ordinate. An artificial neural network (ANN) ensures that optimum colours are placed closest to the sensor by predicting future pointer position.

The CSP successfully provided an input that depicted where a user was pointing. Sensors tested were not able to differentiate more than 28 colours on a TFT screen but an ANN system made them more accurate. Hue and Saturation were successfully used to model colour in a two‐dimensional colour grid as they were unaffected by brightness. The sensor output was successfully converted into a co‐ordinate and used to move a cursor. The kaleidoscope of colours was dynamic as it changed size and colour to increase accuracy. The ANN successfully predicted future positions to assist the sensor in making optimum colour detections. The best predictions of future positions were obtained with four and eight hidden neurons and there did not appear to be any significant advantage in training with more than 2,000 epochs. The system has performed satisfactorily for various inputs.

The novel light pen works on TFT and CRT screens. Problems included the time that colours needed to be left on a screen to be detected and the limited number of colours that could reliably be detected. These were overcome by using a dynamic kaleidoscope of colours to improve colour detection and an ANN to predict future pointer positions, based on previous locations. This prediction was used to place optimum colours closest to the sensor in order to increase accuracy. Further testing is taking place to confirm that the sensor can be used at any angle.

The novel light pen type peripheral and intelligent colour sensing system can replace the need to use a touch screen, mouse or keyboard.

The new intelligent colour sensing system works on either a TFT or CRT screen. The use of the colour translation and sensor and the ANN to predict future pointer position are all novel.