Thesis Defence: Fuel Prediction Modelling for Mining Road Operations

Saira Furqan, supervised by Dr. Warren Hare and Dr. John Thompson, will defend their thesis titled “Fuel Prediction Modelling for Mining Road Operations” in partial fulfillment of the requirements for the degree of Master of Science in Computer Science.

An abstract for Saira Furqan’s thesis is included below.

Defences are open to all members of the campus community as well as the general public. Registration is not required for in-person defences.

---

ABSTRACT

Mining operations depend on the design and optimization of road networks, where trucks move mineral payloads from one location to another. Part of this optimization includes designing roads to minimize truck fuel consumption; however, there is uncertainty surrounding the contribution of road attributes to fuel consumption efficiency. This thesis focuses on analyzing road attributes to predict fuel consumption for mining haul trucks by analyzing a real-time operations dataset.

First, we have developed a four-step methodology for processing and structuring the raw mining fuel operation data into distinct routes. The first step involves cleaning the raw dataset to address missing values and rectify inconsistencies. In the second step, trips are created based on payload data by grouping consecutive records with equal payloads. The third step involves constructing routes by identifying the start-point and end-point of each trip using location data, timestamps, and calculating the distance between these end-points. Lastly, the similarity between routes is calculated using Dynamic Time Warping (DTW) to assess the alignment between different routes.

Following the trip and route construction, we analyze the effects of payload and grade on fuel consumption using multiple linear regression to discover the underlying relationships in the data. We then use non-linear models, including spline and quantile regression, to capture the complex relationships through the median function of the data. These models are evaluated across three types of mining trucks to assess their predictive performance, with results compared to those of an existing nonstochastic inertial model.

The predicted fuel usage shows that each model provides useful information that relates fuel consumption to payload and grade; however, they alone are insufficient to accurately predict fuel consumption amongst changing variability. This highlights the limitations of our model, where incorporating additional road attributes is necessary to improve prediction accuracy under varying operational conditions.