To use AI and machine learning for optimizing manufacturing processes, the first step is to get the data right. The amount and quality of the data collected is key to good data driven modelling of these processes. Thus, it is important to keep track of the quality details when collecting and aggregating the siloed datasets. Data quality management plans are at the core of Industry 4.2^TM and smart manufacturing in the digital transformation age. Data management plans comprise the entire lifecycle of the data from collecting data, integrating the different sources into a coherent format, data normalization and storage to analyzing the data for instant plant action as well as offline.  Data cleansing and pre-processing are pre-requisite to transform the raw data collected into knowledge via such a lifecycle. Due to the diversity in the types of data collected at the foundry, for example, data in different formats, at different levels of granularity and time periods leads to datasets that are siloed, noisy and incomplete in nature.

The focus of our work is on the application of deep learning methods to treat such data quality challenges and to develop data-driven models and obtain insights into the manufacturing processes. This work addresses different dimensions of data quality which are essential for materials processing and manufacturing companies interested in machine learning (as featured in this figure). We provide solutions for these which can be easily incorporated into the existing frameworks of the foundry.  Moreover, the models developed enable the manufacturing companies to do machine learning for applications, such as identifying primary drivers of defect formation, anomaly detection, part quality prediction and predictive maintenance, even when a gold standard data management plan is not readily available.

Click here to learn more about this project and our other groundbreaking research.