To optimize PVC production by precisely predicting the cooling tower's water temperature, we analyzed sensory and weather data from 2015 to 2021, utilizing regression models to understand the factors affecting cooling capacity. This analysis led to a 2-8% increase in PVC production across various lines, employing three models: an updated Excel curve, a linear regression model considering weather parameters, and the most effective LightGBM model incorporating 80 different inputs, including historical data. The implementation of these models significantly improved the accuracy of temperature forecasts, reducing the average error from 1.90 to 0.51, thereby maximizing production efficiency.

To optimise the quality of steel wire by ensuring uniform phosphate coverage, a custom AI solution was built en implemented. By dividing microscope images at a 4000x magnification into several sections, detailed analysis of the phosphate distribution on each part of the steel wire is achieved. ML6 has transitioned from reliance on manual inspection to a data-driven approach, extracting quantitative information from Scanning Electron Microscope (SEM) images automatically. The machine learning models can effectively predict phosphate density, directly from raw SEM images, marking a significant leap forward in precision and efficiency for quality control processes.

Using existing IoT data and building on Google Cloud, ML6 has unlocked significant value, achieving a 200% return on investment within just six months. This approach has had a profound impact on waste reduction and sustainability for our clients, demonstrating the powerful benefits of integrating AI into manufacturing processes.