Using IoT devices and intelligent algorithms to decrease costs and protect the environment

industrial kitchen
Food & Beverage
-
Data science
-
Predictive analytics

The challenge

Cleaning equipment in the food and beverage industry is essential but consumes a lot of water. In view of the drastic environmental changes, our project partner has decided to take extensive measures to improve the energy efficiency, water conservation and resource management of its products and associated processes. By calculating predictions (based on early sensor measurements) on how dirty the equipment is likely to be in advance, the cleaning process can save water in the later stages of the process through decreased water consumption.

Our solution

Solving the problem involved creating and analyzing thousands of time-series derived input features and then selecting the most predictive ones, to make an efficient prediction at each step of the cleaning cycle. This was computationally intensive, and required carefully designing feature calculations, as well as using GPU to solve the high-dimension gradient boosted trees algorithm. Fast computing times allowed a lot of experimentation, which eventually led to the successful model.

Tech stack

Docker
CatBoost
TsFresh

The project

Contiamo was engaged by the project partner after they had successfully developed a technical solution to optimize the cleaning process: by measuring the water pollution during different phases of the cleaning cycle. For each cleaning observation, the data generated consisted of a collection of time series generated by a variety of sensors (flow, pressure, temperature, etc.).

In the first phase of the project, exploratory analytics were performed and trial models were created and presented to the client.

Subsequently, the prediction target was concretely defined. Algorithms were used to generate and test features like turbidity metrics or flow measurements. Based on this, the most promising prediction models were selected.

After the initial training, we realized that a more streamlined and powerful process was required to train and maintain the artificial intelligence. We create a new environment to leverage NVIDA GPUs which significantly reduced the model training time.

After successfully demonstrating the technical feasibility of process optimization, the Contiamo team supported the customer in evaluating the further (organizational and technological) requirements for a real-world implementation. This included setting up the technical infrastructure (real-time sensors) and the data infrastructure as well as an analysis of the return on investment.

Achieved benefits

tick mark

Potential water savings of up to 30%

tick mark

5-fold improvement in prediction precision compared to previous approaches

tick mark

Predictive analytics

The project partner was looking for ways to decrease water consumption during kitchen equipment cleaning by predicting in advance how turbid (or clean) the water would be in the final cycle. Starting with time-series data, Contiamo combined time-series feature engineering with a gradient boosted trees algorithm running on the structured data thus generated. The created artificial intelligence was able to significantly reduce the required water consumption (30%) thanks to its excellent performance.

Are you interested in more information? Let's talk about your current topics and challenges!
Contact