A leading high speed manufacturer, operating in over 36 countries, responsible for supplying many of the brands we know and love today, approached T-DAB.AI with a problem.
They wanted to investigate the power of predictive analytics in their production line but had multiple disparate data sources and minimal internal knowledge.
The manufacturer was keen to understand how machine learning could be applied to reduce machine downtime and spoilage from production errors. The aim was to increase efficiency and productivity, by reducing spoilage through improved stability of the front-end production process, as well as refining material consistency, and enable operations to preemptively intervene.
T-DAB.AI (T-DAB) were engaged to conduct a Discovery & Design exercise and Exploratory Data Analysis (EDA) to assess the data quality and analytical feasibility of developing machine learning applications and outline a high-level architecture diagram to support the solution deployment.
We initially used one years’ worth of data to use machine learning to firstly mine the dataset for key influential features from an initial list of 64 features, and then apply machine learning to predict spoilage and tool failure events within future time periods. Included were machine state, output quality, tool life and operational data.
T-DAB first carried out a data audit, cleaning, and wrangling exercise, followed by feature engineering. Machine learning experimentation was carried out in R.
T-DAB then carried out two exploratory data analyses (EDA) using data from machines, material properties and production. The first of these uncovered insights regarding machine downtime and spoilage, allowing testing of key correlations, and demonstrated the predictive potential of the data for predicting categories of spoilage events, within a defined time period, using relatively coarsely time aggregated data.
The second EDA analysed data from the production of the manufacturing materials used and identified key correlations between material properties, production process features, and material performance (in terms of spoilage and productivity).
Following this work, T-DAB carried out a specification workshop to identify key business challenges faced by the manufacturer, define problem statements, and ideate solutions using analytics and machine learning. This workshop identified multiple machine learning models and outlined the specification for delivery.
T-DAB captured the feedback and analytical roadmap in a PoC report, as well as visualisation and deployment requirements and data platforms required to deploy the algorithms globally.
The project identified immediate opportunities in cost saving for tooling and team training, provided a scalable BI and analytics architecture, and provided recommendations to improve the data collection and visualisation.
In addition, a number of ML algorithms were produced able to predict spoilage and tool failure events to have real world impacts on operational processes in reducing spoilage and downtime.
This resulted in continued project efforts to target £1.5m savings per production line through the application of a global data architecture and algorithm deployment to increase production, material quality and the increasing demand for sustainable manufacturing.