DataProphet Cuts Manufacturing Defects to Zero Using AWS

DataProphet uses AWS to help manufacturers reduce scrap rates to zero. The company optimizes production parameters using machine learning to prevent faulty goods from being produced. It decoupled its AWS environment, with Amazon ECS clusters running machine learning models. Amazon CloudFront delivers the DataProphet web app, which provides an excellent user experience regardless of processing loads elsewhere in the environment.

Most manufacturers identify faulty parts after production and write off the cost. Instead, DataProphet’s PRESCRIBE software uses machine learning (ML) to help customers define the optimum production parameters that prevent faulty parts from ever being made in the first place. Dr. Michael Grant, the company’s chief technology officer, developed DataProphet PRESCRIBE in 2017. “We knew DataProphet PRESCRIBE would have a significant impact on manufacturers’ bottom lines, but we needed a way to deliver it reliably and scale across the world,” he explains.

The software, an expert execution system, constantly receives data from multiple machines in customers’ plants, analyzes it, and delivers results to improve decision-making on the factory floor. One option was to use physical servers to run DataProphet PRESCRIBE, but DataProphet’s customers generate unpredictable loads based on their production cycles. A spike in traffic could exceed the capacity of a fixed hardware environment, meaning clients couldn’t act on the insight that the software generated. DataProphet decided to use Amazon Web Services (AWS) as a platform for product innovation. Grant says, “AWS gives DataProphet PRESCRIBE the flexibility to handle huge spikes in processing loads without affecting the user experience. The instant and continuous insights we give to our clients mean that our customers can adjust machine parameters before defects occur.”

To ensure these instant insights, DataProphet decoupled the major parts of its product. As such, ML models run on Amazon Elastic Container Service (Amazon ECS) clusters, which DataProphet automatically spins up in response to machine requests. The models return inference results, which are then pushed to a web app via an API using Elastic Load Balancing, which shares connections between instances in an automatic-scaling group. The intuitive web-based interface, where manufacturers can see their production data and get their prescriptions, is delivered using Amazon CloudFront. This arrangement remains responsive regardless of the load on other parts of the infrastructure.

Major global manufacturers have been using DataProphet PRESCRIBE since 2017. The solution typically reduces the cost of producing faulty parts by more than 40 percent and has reduced scrap rates to zero for one client. Clients start getting insight from their data in less than three weeks and see a return on their investment in three months on average.

Data protection is vital in manufacturing. By using many unique identity and access controls, encryption keys, and segregating data stores in AWS, DataProphet ensures the protection of each customer’s intellectual property by complete sandboxing. This also means that there’s no danger of data bleeding between clients, which would affect ML models.

Grant concludes, “It’s fair to say that we would not be as successful without the variety, responsiveness, and reliability of the AWS technologies we use.”