The reliability of the data from geotechnical laboratories is an important issue for ensuring the safety and cost-effectiveness of infrastructure design. The DST, although widely used for soil shear strength parameters, is subject to anomalies from equipment drift, operator errors, and digitization problems. Manual review for large datasets is not practical, hence motivating the need for automation in quality control. This paper is concerned with developing and evaluating a complete anomaly detection tool for DST records. It combines a rule-based expert system with a neural network classifier. The workflow includes PDF data extraction, cleaning, feature engineering, and anomaly detection using 996 test records from 69 boreholes of major projects in Bangladesh. For the rule-based system, high transparency and expert agreement at 93% were achieved by flagging obvious anomalies related to implausible densities, water content inconsistencies, and cohesion errors. On the other hand, the neural network revealed adaptability by capturing subtle multivariate anomalies with 72.7% recall but lower precision at 37.2%. In comparative analysis, both approaches proved to possess complementary strengths: rules provide auditability and regulatory compliance, while neural networks increase coverage for complex error patterns. These findings support hybrid frameworks as the most robust solution for digital laboratory transformation, hence enabling scalable, transparent, and reliable geotechnical quality control.
Dey et al. (Wed,) studied this question.