Applying Artificial Intelligence Planning to Optimise Heterogeneous Signal Processing for Surface and Dimensional Measurement Systems

The need for in-process measurement has surpassed the processing capability of traditional computer hardware. As Industry 4.0 changes the way modern manufacturing occurs, researchers and industry are turning to hardware acceleration to increase the performance of their signal processing to allow real-time process and quality control.

This thesis reviewed Industry 4.0 and the challenges that have arisen from transitioning towards a connected smart factory. It has investigated the different hardware acceleration techniques available and the bespoke nature of software that industry and researchers are being forced towards in the pursuit of greater performance. In addition, the application of hardware acceleration within surface and dimensional instrument signal processing was researched and to what extent it is benefitting researchers. The collection of algorithms that the field are using were examined finding significant commonality across multiple instrument types, with work being repeated many times over by different people.

The first use of PDDL to optimise heterogenous signal processing within surface and dimensional measurements is proposed. Optical Signal Processing Workspace (OSPW) is presented as a self-optimising software package using GPGPU acceleration using Compute Unified Device Architecture (CUDA)for Nvidia GPUs. OSPW was designed from scratch to be easy to use with very little-to-no programming experience needed, unlike other popular systems such LabVIEW and MATLAB. It provides an intuitive and easy to navigate User Interface (UI) that allows a user to select the signal processing algorithms required, display system outputs, control actuation devices, and modify capture device properties.

OSPW automatically profiles the execution time of the signal processing algorithms selected by the user and creates and executes a fully optimised version using an AI planning language, Planning Description Domain Language (PDDL), by selecting the optimum architecture for each signal processing function.

OSPW was then evaluated against two case studies, Dispersed Reference Interferometry (DRI) and Line-Scanning Dispersed Interferometry (LSDI). These case studies demonstrated that OSPW can achieve at least21x greater performance than an identical MATLAB implementation with a further 13% improvement found using the PDDL’s heterogenous solution.

This novel approach to providing a configurable signal processing library that is self-optimising using AI planning will provide considerable performance gains to researchers and industrial engineers. With some additional development work it will save both academia and industry time and money which can be reinvested to further advance surface and dimensional instrumentation research.