GRDF : an innovative way to identify chemicals with Cloud AutoML

To drive innovation, we need to do rapid testing directly with the people involved, to develop solutions tailored to real-world issues. We want to find solutions as quickly as possible, and GCP allows us to do that.

About GRDF

With its 200,000 km of pipelines, GRDF (Gaz Réseau Distribution France) manages the largest natural gas network in Europe. Since its creation in 2007, the company has become a dynamic and powerful energy provider in the French utility landscape: in 2017, it had 10.9 million customers and recorded a turnover of €3.6 billion.

The Problem

In order to play a key role in the transition to greener energy, GRDF is implementing important measures such as the development of biomethane and the transformation of the network to increase its efficiency. As part of this transition, the company is considering to expand its offer with services other than gas distribution. For example, it is offering the Gazpar smart meter, a project of unprecedented scale in Europe for the gas network. It also plans to adopt new ways of using data to improve the network in an innovative and agile way.

The Objective

To drive this transformation, GRDF’s digital innovation team decided to find out how machine learning could help them to address their major challenge: ensuring network safety by enabling pipeline technicians and engineers to accurately identify chemicals encountered. 

The Solution

And it was Google Cloud Platform that provided the solution. In addition to developing the application, implementation partner Devoteam G Coud held workshops to explain to the teams of GRDF how the application works and how AI can transform business processes.

“While our data scientists were already familiar with machine learning and TensorFlow, they deepened their knowledge of the technology with great interest,” said Jean-Charles Jorandon.

Before the workshop, our developers’ ML or AI skills were limited. It was really important to us that the team better understand Google’s products and get to work with a real-world use case.”

We chose GCP because the core components of Google Cloud ML Engine are immediately usable. These tools can be mobilised very quickly, so they fit perfectly with the way we test and create prototypes.

Methodology

“Chemical names are often convoluted. We didn’t want the technician to be faced with a long list in which he would have to know the exact name of the product to find it,” explained Jorandon.

With this in mind, GRDF partnered with Devoteam G Cloud to develop an application. The latter uses TensorFlow and Google Cloud ML Engine to identify chemicals based solely on a photograph of their packaging. “Using AI, the app can automatically recognise a product by identifying the shape of the container or its packaging. This allows it to present the correct data sheet to the technician.” added Jorandon.

“For pattern recognition to work, thousands of photographs of the product must be taken. All the chemicals we use were photographed from every angle, distance and configuration. We then made these pictures available on Google Cloud Storage. The Devoteam G Cloud team then defined the parameters on Cloud ML Engine using TensorFlow to create the recognition models. These were then integrated into the application.”

The Result

Thanks to Devoteam G Cloud’s agile development using GCP, the launch of the mobile app took less than two months. “We implemented it in the Lyon area during a three-month test period,” Jorandon said, “and the results are very encouraging: the app recognises 60 products, and allows technicians to easily access safety measures.”

Thanks to Google Cloud, GRDF was able to develop the prototype app quickly, with no unexpected expenses. “The costs are completely transparent and quite reasonable,” Jorandon said. Using GCP allows us to mobilise resources very quickly, at a reasonable cost, in a test scenario.”