Soft sensors based on deep learning regression models are promising approaches to predict real-time fermentation process quality measurements. However, experimental datasets are generally sparse and may contain outliers or corrupted data. This leads to insufficient model prediction performance. Therefore, datasets with a fully distributed solution space are required that enable effective exploration during model training. In this study, the robustness and predictive capability of the underlying model of a soft sensor was improved by generating synthetic datasets for training. The monitoring of intensified ethanol fermentation is used as a case study. Variational autoencoders were employed to create synthetic datasets, which were then combined with original datasets (experimental) to train neural network regression models. These models were tested on original versus augmented datasets to assess prediction improvements. Using the augmented datasets, the soft sensor predictive capability improved by 34%, and variability was reduced by 82%, based on R2 scores. The proposed method offers significant time and cost savings for dataset generation for the deep learning modeling of ethanol fermentation and can be easily adapted to other fermentation processes. This work contributes to the advancement of soft sensor technology, providing practical solutions for enhancing reliability and robustness in large-scale production.
This study presents a novel data-driven modeling approach employing machine learning to develop predictive “soft sensors” for real-time monitoring of ethanol and substrate levels during bioethanol fermentation processes. By utilizing readily measurable parameters such as pH, redox potential, capacitance, and temperature, the model enables continuous prediction of less frequently measured variables including ethanol, substrate, and cell concentrations. Eleven fermentations were conducted, focusing on intensified ethanol production from sugarcane substrate, utilizing cell cycling techniques to augment output. Despite the importance of fermentation data, its acquisition is often constrained by limitations in availability and resources. To address these challenges, this research integrates synthetic time series data generation, thereby enhancing the applicability of machine learning. Through the use of a variational autoencoder (VAE), synthetic time series data was successfully generated, facilitating training and testing of a deep neural network on both original and synthetic datasets. Results demonstrate a significant 30% increase in prediction robustness with the incorporation of generated data, while maintaining comparable accuracy levels. The augmented data effectively enhances the generalization ability of trained models, mitigating overfitting and expanding decision boundaries, thereby overcoming challenges associated with small datasets and inevitable data deviations. This innovative approach offers a promising avenue for enhancing the reliability and scalability of bioethanol fermentation monitoring through AI-based biosensors.