Machine-Learning-Based Screening Tool to Accelerate Aggregate Quality Monitoring in Highway Construction

The Texas Department of Transportation (TxDOT) currently uses the micro-Deval abrasion test, Los Angeles abrasion test, magnesium sulfate soundness (MSS) test, and acid insoluble residue test to monitor the quality of aggregates supplied for TxDOT projects. The MSS test is the most time-intensive, requiring at least 12 business days to sample, test, and report. This study demonstrates the application of a machine learning (ML)-based online tool to accelerate aggregate testing and approval for hot mix asphalt, surface treatment, and Portland cement concrete pavement projects. MSS classifications for aggregates are established based on allowable limits specified by TxDOT. Several ML models have been evaluated to predict MSS values and classifications using data from other quality monitoring tests, covering both TxDOT-approved and new suppliers. A Python-based online interface implements these models to estimate MSS values, provides probabilities of predicted classes, and accepts or rejects aggregates with prediction probabilities of at least 95%. At this probability threshold, the balanced RF algorithm achieves a recall rate of above 97% for mixture-specific binary classification of magnesium soundness across all three applications. An evaluation of the tool’s effectiveness on MSS tests conducted in the laboratory in 2024 indicates that, had decisions been based solely on the predictions of the tool, accepting results with a probability of 95% or higher, the overall approval process could have been expedited by at least 7 business days for approximately 60% of the 1,540 tests completed.