Online Program

Implement Time Series Prediction on Potato Chips Process Control (309880)

Keywords: Keywords: time series analysis, anomaly detection, predictive maintenance, LSTM, XGBoost, TensorFlow, cloud compute, auto-encoder

Digital technologies became the go-to option in recent years for many companies to address the challenges of understanding consumer insights, supply chain operations, and optimize production quality. There are millions of data points being generated by the equipment in the manufacturing sites every second and leveraging the production data can bring in much added value to companies to improve manufacturing efficiency, avoid production down time, and control product quality. This paper discusses how techniques like multivariate time series analysis, anomaly detection, and cloud computing can be used to get sensor data to the cloud and obtain production insights from the sensor data. The goal is not only to identify the abnormal behaviors of the sensor data and the root cause of the abnormal incidents, but also make predictions well in advance so that adjustments can be made proactively and avoid any failures. For this presentation, data were collected from a single manufacturing location and analyzed using emerging machine learning approaches such as LSTM, XGBoost, and Auto-Encoder, to predict the anomaly instance and the corresponding factors that impact production behavior. This work was supported by PepsiCo R&D. The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of PepsiCo, Inc.