View Important Policies and System Requirements for this course
Interested in registering 5 or more engineers for a course? Contact us for information and rates.
INSTRUCTORS:
Jiannan Ding
Chengcheng Tao
Jim Anspach
Mahnaz Rouhi
Purpose and Background
These presentations were recorded at the UESI Pipelines 2025 Conference.
A Machine Learning-Based Interpretable Glass-Box Model for Pipeline Failure Mode Prediction Considering Quantitative Coupling Factors (26 minutes)
This presentation introduced a machine learning–based “glass-box” model designed to predict pipeline failure modes by incorporating diverse and coupled influencing factors. The researchers emphasized that pipeline failures are complex, often resulting from interactions among material properties, operational stresses, and environmental conditions. Unlike traditional black-box approaches, this method provides interpretability, allowing engineers to understand which features drive predictions. Using U.S. crude oil pipeline incident data from 2010–2022, the model integrated spatial, temporal, categorical, and text-based features, ultimately reducing noise and improving prediction reliability. Feature importance analysis showed that pipeline age, pressure margins, and location were among the most influential predictors. The model achieved 90% accuracy, making it both accurate and transparent for operational decision-making.
Visualizing the Z-Value Uncertainty in 3D Utility Models (29 minutes)
This presentation explored the challenges of representing vertical uncertainty (“Z-value”) in 3D utility models. While horizontal location standards such as ASCE 38 have improved plan-view accuracy, vertical positioning remains highly uncertain due to limited survey data, inconsistent records, and varying geophysical methods. The speaker highlighted how uncertainty in depth and shape of buried utilities can undermine applications like clash detection and design coordination. Real-world examples showed how tools like ground-penetrating radar, electromagnetic locators, and as-built standards (ASCE 75) differ in precision and reliability. The discussion emphasized that uncertainty is not just a technical limitation but also tied to data quality, human interpretation, and documentation practices. Ultimately, the presentation advocated for clearer depiction methods, such as probability zones, to better manage risk in utility design and construction.
Sewer Pipe Condition Evaluation Using Ensemble Machine Learning (20 minutes)
This presentation examined how ensemble machine learning techniques can improve sewer pipe condition assessment and predictive maintenance. Traditional approaches such as CCTV inspections are costly, time-consuming, and reactive, prompting the need for data-driven alternatives. By combining decision trees, K-nearest neighbors, logistic regression, and support vector machines, the ensemble model achieved improved predictive accuracy compared to individual methods. The dataset included over 900 sewer segments from Dallas and Tampa, with features such as age, diameter, slope, depth, and material. Results showed that age, material type, and slope were among the most critical deterioration factors, with ensemble voting methods producing an F1 score of 70.6% and AUC values up to 0.96. These insights enable municipalities to prioritize inspections and rehabilitation proactively, maximizing limited budgets and improving infrastructure resilience.
Benefits and Learning Outcomes
Upon completion of this course, you will be able to:
- Explain how interpretable machine learning models can enhance trust and decision-making in pipeline risk assessment.
- Identify key features, such as pipeline age and operating pressure margin, that strongly influence failure mode predictions.
- Discuss the limitations of current methods in accurately capturing Z-value uncertainty for underground utilities.
- Describe how standards like ASCE 38 and ASCE 75 address positional accuracy and where gaps remain in vertical modeling.
- Explain how ensemble machine learning methods improve predictive accuracy in sewer pipe condition assessment.
- List critical pipe features, such as age, material, and slope, that significantly influence deterioration risk.
Assessment of Learning Outcomes
Students' achievement of the learning outcomes will be assessed via a short post-test assessment (true-false, multiple choice, and/or fill in the blank questions).
Who Should Attend?
- Utility and Pipeline Engineers
- Design and Consulting Engineers
- Construction Contractors
- Project Managers
- Academic and Professional Researchers
- Early Career and Pipeline Professionals
How to Earn Your CEUs/PDHs and Receive Your Certificate of Completion
To receive your certificate of completion, you will need to complete a short post-test and receive a passing score of 70% or higher within 1 year of purchasing the course.
How do I convert CEUs to PDHs?
1.0 CEU = 10 PDHs [Example: 0.1 CEU = 1 PDH]