ML dosage predictor to optimise the sterilisation of 1,000 products per week

ML dosage predictor to optimise the sterilisation of 1,000 products per week

Client

STERIS is a leading global provider of products and services that support patient care with an emphasis on infection prevention, focused primarily on healthcare, pharmaceutical and medical device customers, with more than 17,000 associates worldwide.

Background

To optimise its operational processes, STERIS sought to implement an application that could accurately predict ideal sterilisation dosage ranges for products prior to processing.

The solution required a machine learning model capable of forecasting outputs based on key parameters like product density and processing speed. Training on historical data would enable the application to learn optimal dosage ranges to maximise efficiency and minimise risk.

With advanced machine learning capabilities required, the client partnered with Audacia given its track record delivering customised AI solutions.

Solution

Audacia partnered with STERIS to create a machine learning application that could accurately forecast the optimal efficiency of the sterilisation process of thousands of products per week for the UK and US healthcare systems.

Key steps included:

  • Defining the core scenario and prediction requirements
  • Importing and preparing historical data for model training
  • Training a linear regression model using the LGM algorithm
  • Testing the model locally and evaluating performance
  • Enabling retraining as new data becomes available

Trained on 15 years of historic data, the solution uses a linear regression model with LGM algorithms to predict the level of sterilisation required for products prior to product loading based on variables such as belt speed, tote density and adjacent tote exposure. The model is continuously refined with automated retraining, ensuring that predictions remain accurate as new data becomes available.

The application was designed for easy integration with the existing tech stack and ERP platform, with Azure Machine Learning offering added flexibility for training models at scale.

Results

The AI dosage predictor delivered:

97%+ accuracy in forecasting optimal dosage ranges
Faster prototyping compared to traditional ML approaches
Easy integration with existing .NET systems
Flexible training options via local or cloud resources

By leveraging machine learning, the client gained data-driven insights to enhance quality control and maximise productivity.

AI ML case study screenshot