Exploring machine learning to aid the development of manufacturing execution software

Abstract

The rapid development of Internet of Things technology allows cross-communication between modern machines via a common cloud platform. This technology has increasingly been adopted in manufacturing environments giving rise to what we know as Industry 4.0. Manufacturing environments that were successful in adopting the new technology in their production systems are often referred to as Cyber-Physical Production Systems. Cyber-Physical Production Systems are built on existing production monitoring platforms such as the Manufacturing Execution Software and go a step further to connect the data to a cloud environment. Allowing the manufacturing environment to have complete control and analytic capability of their smart machinery over the cloud. Improvements over the state of existing Manufacturing Execution Software are key to the adoption Cyber-Physical Production Systems. However, it is necessary to improve the existing software services to meet the inherent design criteria for the adoption of Cyber-Physical Production Systems. With the onset of Industrial 4.0, there is increasing demand for the development of better Manufacturing Execution Software to better support the integration Cyber-Physical Production Systems. This surge in demand prompts the need for a quicker development cycle for Manufacturing Execution Software. In the development of bespoke Manufacturing Execution Software for individual production plants, a large amount of qualitative feedback data was gathered from the manufacturing operators and the software implementers. This poses an issue for software implementors as they are required to expend significant time analysing a large volume of qualitative data to generate meaningful developmental insights. This process can be expedited if we can classify the qualitative feedback data into priorities allowing developmental efforts to be better focused on meaningful critical developments. That can be achieved using Natural Language Processing to extract information from unstructured qualitative feedback data and classify them based on their attributes using a Machine Learning model. This task had been replicated in various applications to conduct sentiment analysis where text is classified based on the emotional sentiments that it demonstrates, and also in-text classification to sort a large volume of information into various categories. This project aims to evaluate the capabilities of Machine Learning in classifying qualitative feedback data of Manufacturing Software. A binary classification method is adopted to predict the feedback data into two categories, feedback involving improvement requests and feedback not involving improvement requests. This approach constitutes the usage of classification algorithms and neural networks. Deep learning and classification are possible with TensorFlow as TensorFlow is capable of building neural networks used for Classification, Perception, and Prediction. GloVe word embedding was used for word vectorisation, word vectorisation is important to present unstructured data in vectors allowing the Machine Learning model to mathematically compute their relationship with each other. The Keras library is used to manage and operate TensorFlow as it is an API that allows a more user-friendly approach due to its nature as a high-level API.