Maïeul Lombard

Zaack offers an indoor air quality monitoring solution. Zaack has a park of 500 sensors installed worldwide that transmit concentration measurements every 30 seconds.

Services:

• Diagnosis: identification of issues, solution audit, analysis of data assets and processes

• Framing: identification of data science use cases and definition of the development roadmap: sensor measurement reliability, automation of report and alert generation for clients, anomaly detection, and predictive model development

• Reliability of electrochemical sensor measurements (NO2, Ozone, CO):

o Analysis of correlations with Python between raw data (electrical potentials for each pollutant), temperature, humidity, and reference data obtained from analyzers

o Development of an experimental plan to define a reliable statistical model of pollutant concentration from raw data

o Development of a calibration procedure for new sensors (Python script)

• Automation of report generation and pollution threshold alert processes: development of Python scripts deployed between the ETL dataflow and the MySQL database

• Anomaly detection:

o Exploratory data analysis of about a hundred installed sensors to characterize anomalies (sensor drifts, high pollution peaks, sensor malfunction anomalies, etc.)

o Use of statistical methods (winsorization and moving average) to remove signal noise

o Development of a method based on stationarity tests to detect sensor drifts

o Training of autoencoders to detect anomalies

o Validation of the approach on installed sensors

• Predictive modeling:

o Analysis of the correlation between external data (weather and outdoor air quality) and Zaack sensor data

o Development of a method for extrapolating external data from the nearest stations to the installed sensor

o Development of predictive models using sensor data and forecasts for external data on test sensors

Field survey and needs analysis of catering staff, dietitians, the logistician, and the production manager

Description of the current process (from menu planning to meal delivery) and identification of decision-making points that can be facilitated by statistical algorithms

Formalization of the AI specifications and the system to improve the meal ordering process

Collection of historical data (production IS) and Excel files previously used in the Information System

Data cleaning and consistency check

Descriptive and inferential statistical analyses to identify factors that most influence school canteen attendance

Statistical modeling of school canteen attendance (Machine Learning) -> reduced overproduction waste by 25% to 45%