George-Christopher Vosniakos

Optimizing feedforward artificial neural network architecture

Despite the fact that feedforward artificial neural networks (ANNs) have been a hot topic of research for many years there still are certain issues regarding the development of an ANN model, resulting in a lack of absolute guarantee that the model will perform well for the problem at hand. The multitude of different approaches that have been adopted in order to deal with this problem have investigated all aspects of the ANN modelling procedure, from training data collection and pre/post-processing to elaborate training schemes and algorithms. Increased attention is especially directed to proposing a systematic way to establish an appropriate architecture in contrast to the current common practice that calls for a repetitive trial-and-error process, which is time-consuming and produces uncertain results. This paper proposes such a methodology for determining the best architecture and is based on the use of a genetic algorithm (GA) and the development of novel criteria that quantify an ANNs performance (both training and generalization) as well as its complexity. This approach is implemented in software and tested based on experimental data capturing workpiece elastic deflection in turning. The intention is to present simultaneously the approachs theoretical background and its practical application in real-life engineering problems. Results show that the approach performs better than a human expert, at the same time offering many advantages in comparison to similar approaches found in literature.

Recognizing D shape features using a neural network and heuristics

This work presents a Feature Recognition system developed using a previously trained Artificial Neural Network. The part description is taken from a B-rep solid modellers data base. This description refers only to topological information about the faces in the part in the form of an Attributed Adjacency Graph. A set of heuristics is used for breaking down this compound feature graph into subgraphs, that correspond to simple features. Special representation patterns are then constructed for each of these subgraphs. These patterns are presented to a Neural Network which classifies them into feature classes: pockets, slots, passages, protrusions, steps, blind slots, corner pockets, and holes. The scope of instances/ variations of these features that can be recognised is very wide. A commercially available neural network modelling tool was used for training. The user interface to the neural network recogniser has been written in Pascal. The program can handle parts with up to 200 planar or curved faces. The performance of the recogniser in terms of speed is far better than that of any other rule-based system due to the Neural Network approach employed. The basic limitation is that of the heuristics used to break down compound features into simple ones which are fed to the ANN, but this is still a step ahead compared to other approaches.

An IGES post-processor for interfacing CAD and CAPP of 2 1/2D prismatic parts

An IGES post-processor is presented, written in FORTRAN-77, to extract from an IGES file all information that an automatic process planning program needs, derive new information when necessary and formalise it as PROLOG facts. The processor deals with wire frame models and drawings. Its various modules are described and considerations about its design are discussed, associated with the IGES specification philosophy and pre-processor implementations. Analogous approaches from other processor designs are reviewed.

Offline flexible optimisation of feed and speed in computer numerical control machining of sculptured surfaces exploiting dedicated cutting force metamodels

An intelligent offline optimisation system for feed and spindle speed in three-axis machining of sculptured surfaces is presented. Optimisation is performed by a standard genetic algorithm seeking, in this instance, to minimise machining force variation along designated sections of the tool path. Machining force is modelled realistically, rather than mechanistically, by dedicated artificial neural networks trained with a low number of experimental data that are obtained from the very equipment on which machining is to be carried out. The artificial neural networks employ the material volume removed per tool revolution as calculated using a standard solid modelling engine. The artificial neural network model architecture is optimally determined in a rigorous way as opposed to often followed trial-and-error approaches, and it is made use of directly in the evaluation function of the optimiser. This framework is flexible enough to enable definition of constraints, additional evaluation criteria and desirable force patterns to guide optimisation. Effectiveness of the approach is successfully demonstrated for finish machining a hemispherical part.

Knowledge-based automatic allocation of cad drawing annotation to wireframe iges models of 2 1/2-D prismatic parts

A knowledge-based Prolog program is presented that interprets the annotation information of a CAD drawing in terms of geometry primitives derived from the wireframe model used to create the drawing. The knowledge base consists of rules for recognising drawing symbols, parsing annotation text, identifying the connecting elements between text and geometry, suggesting geometry primitives to be sought for different annotation entities, and defining geometric connectivity relations. This enables the knowledge base to be incrementally developed to cater for more cases. So far, size tolerances, surface finish and some drawing notes are dealt with by the system. All the information needed as input is obtained from the components IGES file.

Beware of the Robot: A Highly Interactive and Immersive Virtual Reality Training Application in Robotic Manufacturing Systems

A highly interactive and immersive Virtual Reality Training System (VRTS) is developed, in terms of an educational serious game that simulates the cooperation between industrial robotic manipulators and humans, executing manufacturing tasks. “BeWare of the robot” application ultimately aims at studying the acceptability of human-robot collaboration, when both human and robot share the same workspace. The initial version of the application was evaluated by a group of users. Experimental results on usability and technical aspects are presented and several remarks about users’ experience and behavior in the virtual world are discussed.

Cognitive task analysis for virtual reality training: the case of CNC tool offsetting

Motivation -- To examine if cognitive task analysis of expert machinists can be effective in developing a virtual reality based training system for CNC tool offsetting.n Research approach -- A cognitive task analysis of expert machinists was conducted which informed the development of a VR training system for CNC tool offsetting. Subsequently the effectiveness of the analysis was evaluated by conducting an experiment with 31 mechanical engineering students.n Findings/Design -- The virtual reality system demonstrated positive training transfer for the task of tool offsetting. The above indicates that the cognitive task analysis performed was effective in identifying a number of key skills of the tool offsetting task.n Research limitations/Implications -- The study does not prove the superiority of cognitive task analysis over other approaches for specifying virtual reality training systems, since it does not compare the cognitively tuned system with another one.n Originality/Value -- The present work provides evidence that skill transfer can be achieved even with low physical fidelity provided that the cognitive organization of a task is adequately mapped in the virtual reality system.n Take away message -- Further and beyond fidelity issues, cognitive task analysis can provide important input in specifying effective VR training systems.

Modelling simple human-robot collaborative manufacturing tasks in interactive virtual environments

This paper presents in brief a novel interactive Virtual Environment (VE) that simulates in real-time collaborative manufacturing tasks between a human and an industrial robotic manipulator, working in close proximity, while sharing their workspaces. The use case scenario is highly collaborative and incorporates a wide variety of interaction tasks, such as: collaborative handling, manipulation, removal, placement and laying of carbon fabric composite parts. A Kinect sensor and a Head Mounted Display (Oculus Rift) are employed as 3D User Interfaces for interaction, immersion and skeletal tracking of the user motion. In this paper, particular emphasis is given to the various interaction techniques used to facilitate implementation of virtual Human-Robot Collaboration (HRC). The collaborative tasks are principally executed with contactless, natural and direct interaction. In addition, two novel interaction metaphors were developed. The real fabric laying task and the backing film removal task are reproduced in the VE with the implementation of the follow-my-hand technique; the user has to follow with his hand a virtual hand-like index (guide) that moves along a predefined pattern. Preliminary findings concerning the effectiveness of HRC modelling tasks are positive, and are briefly discussed.

Intelligent Optimisation of 3-Axis Sculptured Surface Machining on Existing CAM Systems

This chapter discusses how the current practice for sculptured surface machining may be improved by embedding intelligent optimisation in the process planning steps for both roughing and finishing. Complexity and interdependencies between machining parameters are dealt with by stochastic evolutionary techniques whilst the need to take into account results of calculations lead to embedding meta-models of the machining process. Formulations are provided for the optimisation criteria including machining time, remaining material volume, distribution of remaining material volume, cutting force variation etc. and examples are given for considering machining constraints. Integration with existing CAM systems APIs is discussed and the possible future trends are presented.

On reference modelling of integrated manufacturing systems using OOA

This work advocates the use of Object Orientation as basis of a methodology for creating reference models of manufacturing systems. Work on reference modelling of manufacturing systems integration is reviewed first. Then this work’s own standpoint on reference modelling is outlined with definition of the main concepts. Reference models are distinguished from actual-implementation models through the need for generalisation, abstraction and expansion ability. Coad and Yourdon’s Object Oriented Analysis is used to demonstrate reference modelling of structural, transformation and procedural aspects of a batch-processing shop-floor system, explaining how manufacturing systems entities are mapped to OOA entities. The main concepts introduced along with discussion of relevant examples are a phenomenological View, application-related Views, abstraction hierarchy of objects and expansion ability of the model to advance from abstract to environment-specific reference models. Modular instancing of entities is also discussed in the context of further work.