Case study

Trend Anomaly Detection

Advanced analysis of microbiological tests for the pharmaceutical sector

Client: Pharmaceutical Multinational

Introduction

The reference client for this solution is a multinational operating in the chemical/pharmaceutical sector with business processes that include water and air quality testing.

The need was to automate and streamline the microbiological analysis process, often complex and resource-intensive in terms of time and effort.

For this reason, we developed an AI-based solution to detect anomalies and unwanted trends in environmental monitoring data.

Challenges

Challenges Addressed

Microbiological testing of water and air at a pharmaceutical company is normally affected by several issues:

High Costs

Given the criticality of the sector, errors in analysis can translate into significant costs for the company.

Significant Complexity

The analysis procedure normally requires expert and qualified personnel.

Long Time

The entire analysis procedure takes a long time to complete.

Goals

Project Goal

Build an analysis tool for microbiological water and air tests that can improve the analysis process by making it:

Automatic

The operator no longer has to personally review all results.

Continuous

The system acquires and analyzes data 24/7.

Fast

Analysis is performed by a modern computing system, much faster than a human operator.

AI solution

Our Solution

Build an environmental test data analysis system capable of:

Learning from Data

Through advanced Machine Learning algorithms, the system can automatically adapt to the data and problems presented to it.

Detecting Anomalies

The system can automatically identify anomalous data and flag them to the expert operator.

Roadmap

Implementation

The tool was built by combining several Machine Learning strategies:

Data Synthesis and Augmentation

In addition to building a dataset by collecting environmental monitoring data from the company (with corresponding anomaly labels for the system to predict), we used algorithms to generate synthetic samples to increase dataset size and train the models more effectively.

Supervised Learning

The resulting dataset was used to train supervised classification models.

Model Ensemble

To further increase system accuracy, we developed policies to manage the outputs of different Machine Learning models trained on the same dataset.

Results & Benefits

Results and Benefits

Concrete Results

The resulting system can detect anomalous and unwanted trends in the data with over 90% accuracy.

Benefits Achieved

Cost Reduction

Several man-hours of expert operators are no longer needed to conduct the analyses, leading to significant savings.

Objective Analyses

Results are more objective as they no longer depend on subjective interpretations.

Conclusion

Conclusions

The developed solution proved highly effective, enabling significant savings in terms of working hours and time, and improving the quality of the analyses performed.

Thanks to the use of advanced Machine Learning and data analysis technologies, the solution can also be easily extended and adapted to other types of environmental analysis.

Case studies