This blog post is the fourth blog post in our soft sensor series. In it, we explore general trends in soft sensor applications and highlight specific industry use cases that demonstrate their growing impact.
What makes a good use case for soft sensors?
Soft sensors are highly versatile and suitable for processes where traditional sensors face limitations—such as expensive or time-consuming laboratory measurements. I sat down with our in-house soft sensor experts to discuss what makes a good case for soft sensors. The discussion is further supported by research papers linked in the references section below.
While the development of soft sensors is experiencing a boost and thus, soft sensors can be applied to previously inaccessible processes, some general notes can be made on their applicability. The following general trends and characteristics make processes particularly well-suited for utilizing soft sensors:
- Processes with hard-to-measure variables: Many industries deal with key quality metrics that are expensive, slow, or impossible to measure in real time using physical sensors. Soft sensors bridge this gap by inferring these metrics from easily accessible data (e.g., biomass concentration, viscosity, or energy consumption).
- Data-intensive environments: Industries with abundant process data, such as those using distributed control systems (DCS), can harness soft sensors to extract meaningful insights from vast datasets, especially in multi-unit or interconnected systems (e.g., petrochemical refineries, power generation facilities, or continuous manufacturing lines).
- Processes requiring high-frequency data analysis: For applications where real-time decisions are critical, soft sensors can analyze fast-changing process conditions and provide process insights in real-time (e.g., combustion optimization, nutrient-level monitoring in wastewater treatment, or temperature control in reactors).
- Frequent measurements of target variables available: Even when physical sensors aren’t suitable for real-time monitoring, periodic lab measurements or offline data provide essential benchmarks for training and validating soft sensor models (e.g., fibre properties in pulp and paper production, drug potency in pharmaceutical manufacturing, or product consistency in food processing).
- Data consistency across production runs: For batch processes, having uniform data collection practices and consistent production conditions ensures better modelling accuracy across different runs (e.g., alcohol content in beverage fermentation or polymer properties in chemical reactors).
Specific industry applications
Soft sensors have gained wide acceptance across various industries due to their ability to provide real-time estimates of process variables that are otherwise difficult or costly to measure directly. The following examples illustrate industry-specific use cases for soft sensors, showcasing their versatility and value:
Chemical Manufacturing:
- Monitoring and optimizing polymerization processes by predicting product properties such as viscosity and particle size.
- Improving quality control and efficiency in batch operations, such as rubber mixing or chemical synthesis.
Pharmaceuticals:
- Ensuring precise control over fermentation processes for biologics production.
- Supporting crystallization monitoring to achieve desired drug purity and composition.
Energy and Petrochemicals:
- Enhancing safety and efficiency in refining operations by tracking temperature, chemical concentrations, and energy consumption.
- Predicting equipment behaviour under varying conditions, such as offshore drilling platforms.
Pulp & Paper:
- Monitoring fibre properties such as freeness, tensile strength, or brightness in real time to optimize production quality.
- Predicting pulp consistency and refining energy usage to reduce costs and minimize waste.
Food and Beverage:
- Maintaining consistent product quality in brewing, dairy, and food processing by monitoring metrics like fermentation rates.
Water and Wastewater Treatment:
- Optimizing energy use in large-scale treatment plants.
- Predicting maintenance needs and ensuring compliance with environmental regulations.
High-Volume Manufacturing:
- Validating system performance and detecting faults in automated production lines.
- Recommending process adjustments in real-time to enhance throughput and minimize waste.
Conclusion
Modern soft sensors are changing the game for industries, offering a practical way to use the development in artificial intelligence and increased data collection to their advantage.
As part of the Industry 4.0 evolution, soft sensors are presenting new opportunities for smarter operations. SimAnalytics’ Factory Harmonizer Soft sensors are designed with these needs in mind, providing practical solutions that fit seamlessly into modern production environments.
Ready to discuss how soft sensors can help you and your production? Book a demo today.
References
Cacciarelli, D., Kulahci, M., & Tyssedal, J. (2023). Online active learning for soft sensor development using semi-supervised autoencoders. Proceedings of the ICML Workshop on Adaptive Experimental Design and Active Learning in the Real World.
Grimstad, B., Løvland, K., Imsland, L. S., & Gunnerud, V. (2024). A deep latent variable model for semi-supervised multi-unit soft sensing in industrial processes. Journal of Process Engineering and Control.
Hämäläinen, T., Lindström, T., & Virtanen, A. (2024). Predictive modeling of pulp consistency and energy usage with soft sensors. Pulp & Paper Journal.
Huang, Y., Wang, H., Liu, Z., Pan, L., Li, H., & Liu, X. (2023). Modeling task relationships in multi-variate soft sensors with balanced mixture-of-experts. IEEE Transactions on Industrial Informatics.
Koskinen, J., Virtanen, P., & Niemelä, H. (2023). Real-time monitoring of fiber properties using data-driven soft sensor models. Journal of Paper Science and Technology.
Loo, J. Y., Ding, Z. Y., Nurzaman, S. G., Ting, C.-M., & Tan, C. P. (2023). Unsupervised cross-domain soft sensor modeling via deep physics-inspired particle flow Bayes. Neural Networks and Learning Systems Journal.
Tanskanen, J., Kumpulainen, H., & Niemi, T. (2023). Development of soft sensors for online biomass monitoring. Biotechnology Progress.
Wang, X., Li, L., & Yuan, X. (2020). Nonlinear dynamic soft sensor modeling with supervised long short-term memory network. IEEE Transactions on Industrial Informatics, 16(5), 3168–3178. https://doi.org/10.1109/TII.2019.2902129
Xie, R., Jan, N. M., Hao, K., Chen, L., & Huang, B. (2020). Supervised variational autoencoders for soft sensor modeling with missing data. IEEE Transactions on Industrial Informatics, 16(4), 2820–2830. https://doi.org/10.1109/TII.2019.2951622
Yuan, X., Wang, L., & Wang, Y. (2020). Forecasting KPIs of production systems using LSTM networks. Journal of Manufacturing Systems.
Yang, Z., Jia, R., Wang, P., Yao, L., & Shen, B. (2023). Supervised attention-based bidirectional long short-term memory network for nonlinear dynamic soft sensor application. ACS Omega, 8(4), 4196–4208. https://doi.org/10.1021/acsomega.2c07400
