Read more
Neural network technology encompasses a class of methodswhich attempt to mimic the basic structures used in thebrain for information processing. Thetechnology is aimed atproblems such as pattern recognition which are difficult fortraditional computational methods. Neural networks havepotential applications in many industrial areas such asadvanced robotics, operations research, and processengineering.This book is concerned with the application of neuralnetwork technology to real industrial problems. Itsummarizes a three-year collaborative international projectcalled ANNIE (Applications of Neural Networks for Industryin Europe) which was jointly funded by industry and theEuropean Commission within the ESPRIT programme. As a recordof a working project, the book gives an insight into thereal problems faced in taking a new technology from theworkbench into a live industrial application, and shows justhow it can be achieved. It stresses the comparison betweenneural networks and conventional approaches. Even thenon-specialist reader will benefit from understanding thelimitations as well as the advantages of the new technology.
List of contents
1 Introduction.- 1.1 Purpose of the handbook.- 1.2 Origins of the ANNIE project.- 1.3 The ANNIE team.- 1.4 Overall objectives of the ANNIE project.- 1.5 Applications selected for demonstration of neural network capability.- 1.6 Relationship to ESPRIT aims and objectives.- 1.7 Layout of the handbook.- 2 An Overview of Neural Networks.- 2.1 The neural network model.- 2.2 Principal features.- 2.3 Neural networks used in ANNIE.- 3 Implementations of Neural Networks.- 3.1 Sequential implementation.- 3.2 Examples of implementations of neural networks.- 3.3 Parallel implementation.- 3.4 Discussion.- 3.5 Hardware.- 3.6 Floating point systems.- 3.7 New processors and components.- 3.8 Systolic computation.- 3.9 Summary of architectural features.- 3.10 Benchmarking.- 3.11 Software.- 3.12 Environments developed within ANNIE.- 3.13 Dedicated neural network hardware.- 4 Pattern Recognition.- 4.1 Introduction.- 4.2 Learning mechanisms and evaluation criteria.- 4.3 Generic problems identified by the partners.- 4.4 Supervised learning on generic datasets.- 4.5 Unsupervised learning.- 4.6 Applications of neural networks to pattern recognition in acoustic emission.- 4.7 Proof testing of pressure vessels.- 4.8 Detection and characterisation of defects in welds from ultrasonic testing.- 4.9 ALOC defect detection.- 4.10 Solder joints inspection with neural networks from 3D laser scanning.- 4.11 Conclusions.- 5 Control Applications.- 5.1 Introduction.- 5.2 Overview on control technology.- 5.3 Use of neural networks for control purposes.- 5.4 Lernfahrzeug system (NeVIS).- 5.5 NeVIS IV.- 5.6 Methodology.- 5.7 Identification of a moving robot.- 6 Optimisation.- 6.1 Introduction.- 6.2 Conventional methods in combinatorial optimisation.- 6.3 Linear programming.- 6.4 Integer linear programming.- 6.5 Heuristics.- 6.6 Neural network methods in combinatorial optimisation.- 6.7 The crew scheduling problem.- 6.8 A specific airline case.- 6.9 The pairing generator.- 6.10 Conventional methods for set covering problems.- 6.11 The neural network approach.- 6.12 Improving the performance of the network.- 7 Methodology.- 7.1 Introduction.- 7.2 Conventional and neural network approaches.- 7.3 Implementing solutions.- 7.4 ANNIE applications.- 7.5 Discussion.- Appendix 1: Partners in the ANNIE Consortium and Project Staff.- Appendix 2: Networks Used in the Project.- A2.1 Introduction.- A2.2 Associative networks.- A2.3 Linear associative networks.- A2.4 Hopfield networks.- A2.5 Bidirectional associative memories.- A2.6 The Boltzmann machine.- A2.7 Error feedback networks.- A2.8 Error feedback learning.- A2.9 The back-propagation algorithm.- A2.10 Self-organising networks.- A2.11 Further studies.- Appendix 3: ANNIE Benchmark Code.- A3.1 Introduction.- A3.2 Interpretation of benchmarks.- A3.3 Some results.- A3.4 Test Code.- Appendix 4: Some Suppliers of Network Simulators.
