A collaborative approach allows smaller enterprises to access smart industrial technology

What is it about?

Smart industries have been able to streamline much of their operations using machine learning, smart sensors, and cloud technology. How do they do this? Smart industries host machine learning algorithms in their centralized servers. These algorithms collect data from the industry's processes. Then, they analyze this data. Their output includes ways to boost production, reduce waste, and save energy for the industries. However, the algorithms are built and trained on a large data set first. The problem? Smaller industries lack these datasets. Because of this, the benefits of smart industries have remained out of reach for them. This paper presents an approach where different manufacturers can combine their datasets to build a shared model while still keeping their sensitive information private. This approach of keeping local data separate is called federated learning. It's better for privacy! Here, an initial version of the model is shared with all the participants. Each manufacturer then trains the model on their private network and sends the results back to the central server. Their data is then used to improve the shared model. This approach can be used for processes that are common across different manufacturing plants. For example, the study used it to build models that could predict defects in sheet metal forming.

Featured Image

Photo by Markus Winkler on Unsplash

Photo by Markus Winkler on Unsplash

Why is it important?

Sheet metal forming is commonly used to make a variety of items. These can range from highly complex aircraft wings to simpler items like vehicle doors and food cans. Thus, the approach presented in this study can be used in a broad range of industries. It would help them improve their processes based on shared data, which means they have a larger dataset to learn from. KEY TAKEAWAY: In the proposed approach, industries share data to build a machine learning model. This model is then trained locally. This way, a larger dataset is obtained to train and build the model. Moreover, the restrictions of building a model from the ground up are removed. The privacy of the training data is ensured too. This method allows smaller firms to take advantage of machine learning techniques. Keywords/meta tags: