Data-Driven Soft Sensor Deployment in a Tar Distillation Plant: A Strategy for Stepwise Digitalization

Digital transformation in established chemical facilities is often hindered by infrastructure gaps, unclear return on investment, and a lack of organizational readiness. While data-driven soft sensors offer significant potential for real-time quality monitoring, their adoption is often constrained by systemic limitations. This paper presents an implementation strategy for a soft sensor at an industrial tar distillation facility to estimate the pitch softening point. Rather than treating technical and organizational limitations as blockers, they are treated as design boundaries that define a stepwise deployment roadmap. A baseline linear model was developed to run within the existing SCADA system, achieving a prediction accuracy of 58% within tolerance. This provides a zero-cost early alarm system for production consistency without requiring new hardware. The study further identifies the specific technical infrastructure requirements, including data pipelines and memory modules, needed to transition to more advanced nonlinear models, thereby achieving 92% accuracy. Following deployment of the baseline, the facility elevated the identified constraints and now operates a nonlinear, data-driven soft sensor for process monitoring, which has been tested in closed-loop control. The approach enables stepwise, sustainable digitalization by aligning technical development with the infrastructure investment required for digital maturity.