Keywords: Machine learning
Abstract: Stone Three is an Industrial Internet-of-Things company that develops AI-augmented solutions within the digital productivity, workplace safety and employee healthcare sectors. In the digital productivity field. They offer smart sensors that leverage machine vision technology, as well as offering process monitoring services: generating actionable advisories for improved process performance. Stone Three has been providing state-of-the-art computer vision smart sensors to the mining industry for more than 20 years. In this talk, Stone Three will share their insights on the challenges and best practices for scalable machine vision smart sensors: including what is required for industrial image data collection, image labelling, label review, model training and review, model deployment and model updates.

Keywords: Machine learning

Keywords: Machine learning
Abstract: Real time analyses of the composition of intermediate or final product streams are often not available. This can be due to initial capital optimisation or even measurement practicalities. Continuous optimisation of such processes then require the estimation of the key components that constrain profitability at suitable intervals. In this workshop we demonstrate the application of industrial statistics and data science to derive a model for estimating a key chemical component in a product stream. The derived soft sensor is then employed in a closed loop control strategy. The aim of the control strategy is to reduce the standard deviation of the constraining variable and to shift it’s mean operating point closer to the constraint in order to optimise the process. We will demonstrate how data science techniques can be used to develop such models and how a non-perfect, but directionally correct, model can be applied in practice to improve production efficiency.

Keywords: Machine learning

Keywords: Machine learning
Abstract: A process health index is a key performance indicator in industrial processes for detecting abnormal behaviour and to performing diagnostic analysis in order to make the necessary modifications to the processes. However, performance measures in industrial processes are functions of many variables which are commonly operated in very narrow ranges, which makes the quantification of the multivariate relationships challenging. In this paper we discuss how Machine Learning algorithms are employed to quantify the relationships between performance measures and multivariate inputs. In addition, given the input-output relationships, we present how a multivariate process health index can be specified for multivariate process monitoring and diagnostic analysis. We will illustrate how these methods have been applied successfully on some commercial processes.

Keywords: Machine learning

Keywords: Machine learning
Abstract: Boilers are critical equipment used for steam generation and are found throughout the manufacturing industry. Unplanned downtime of boilers can lead to a halt in production as steam supply often fills a process-critical function. In this workshop we will demonstrate how an intelligent soft sensor was developed using Machine Learning (ML) for a customer in the sugar industry. Steam demand within this particular sugar mill fluctuate drastically and disturbs the boiler up to a point where trips occur due to extreme water levels. The soft sensor algorithm quantifies the likelihood of a boiler trip and predicts when the next trip would occur. This information is used by operators to take appropriate actions in advance to prevent possible trips from happening. The soft sensor could possibly be used within Model Predictive Control in future to automate intervention.

Keywords: Machine learning