This work describes the collection and properties of the publicly available rock drill fault classification data set rockdrill11, used for the 2022 PHM Conference Data Challenge. The data is collected from a carefully instrumented hydraulic rock drill, operating in normal operation in a test cell while inducing a number of faults. Hydraulic pressure is measured at 50kHz at three different locations, resulting in detailed pressure signatures for each fault. Due to wave propagation phenomena, the system is sensitive to individual differences between different rock drills, drills rigs and configurations. Such differences named "individuals" are introduced in the data by altering certain parameters in the test setup. An important part of the data is therefore the availability of No-fault reference cycles, which are supplied for all individuals. These reference cycles give information on how individuals differ from each other, and can be used to improve classification.

The global mining industry is currently facing a huge transition from manually operated individual vehicles, to autonomous vehicles being part of an industrial process-like environment. The change is driven by the never ending need for efficient, safe, and environmentally friendly operations. One intentional consequence is an increased distance between the operator, and the machine being operated. This enables safer working environments and reduced cost for ventilation and other supporting systems in a mine, but it also results in the loss of the systems most important sensor. The transition from manual to autonomous operation requires this gap to be filled from a system awareness perspective, which lately has become evident with the large resources that car manufacturers use to develop self-driving cars. This thesis also targets system awareness, but of the internal kind. By this we mean knowing the condition of the machine and its capabilities. The operator is the most important sensor also for internal condition, and if no operator is present on the machine, this gap needs to be filled.

The mining industry is categorized by small series and significant customization of machinery. This is a direct result of the geological prerequisites, where differently shaped ore bodies cause large differences in mine layout and mining methods. This thesis explores how methods estimating the health of mining vehicles can be used in this setting, by utilizing sensor signals to make assessments of the current vehicle condition and tasks.

The resulting health information can be used both to aid in tasks such as maintenance planning, but also as an important input to decision making for the planning system, i.e. how to run the vehicle for minimum wear and damage, while maintaining other mission objectives.

Two applications are studied. Mine trucks have slow degradation modes, such as crack propagation and fatigue, that are difficult to handle with data driven approaches since data collection requires significant amounts of time. A contribution in this thesis, is a method to utilize short term measurement data together with data driven methods to obtain the loads of a vehicle, and then to use physics based approaches to estimate the actual damage.

The second application considers monitoring faults in hydraulic rock drills using online measurements during operation. The rock drill is a specifically difficult case, since severe vibration levels limits the locations and types of sensors that can be used. The main contribution is a method to handle individual differences when classifying internal faults using a single pressure sensor on the hydraulic supply line. A complicating factor is the large influence of wave propagation, causing different individuals to show different behavior.

This work presents a method for on-line condition monitoring of a hydraulic rock drill, though some of the findings can likely be applied in other applications. A fundamental difficulty for the rock drill application is discussed, namely the similarity between frequencies of internal standing waves and rock drill operation. This results in unpredictable pressure oscillations and superposition, which makes synchronization between measurement and model difficult. To overcome this, a data driven approach is proposed. The number and types of sensors are restricted due to harsh environmental conditions, and only operational data is available. Some faults are shown to be detectable using hand-crafted engineering features, with a direct physical connection to the fault of interest. Such features are easily interpreted and are shown to be robust against disturbances. Other faults are detected by classifying measured signals against a known reference. Dynamic Time Warping is shown to be an efficient way to measure similarity for cyclic signals with stochastic elements from disturbances, wave propagation and different durations, and also for cases with very small differences in measured pressure signals. Together, the two methods enables a step towards condition monitoring of a rock drill, robustly detecting very small changes in behaviour using a minimum amount of sensors. Copyright (C) 2021 The Authors.

Poor road conditions in underground mine tunnels can lead to decreased production efficiency and increased wear on production vehicles. A prototype system for road condition monitoring is presented in this paper to counteract this. The system consists of three components i.e. localization, road monitoring, and scheduling. The localization of vehicles is performed using a Rao-Blackwellized extended particle filter, combining vehicle mounted sensors with signal strengths of WiFi access points. Two methods for road monitoring are described: a Kalman filter used together with a model of the vehicle suspension system, and a relative condition measure based on the power spectral density. Lastly, a method for taking automatic action on an ill-conditioned road segment is proposed in the form of a rescheduling algorithm. The scheduling algorithm is based on the large neighborhood search and is used to integrate road service activities in the short-term production schedule while minimizing introduced production disturbances. The system is demonstrated on experimental data collected in a Swedish underground mine.

The life and condition of a mine truck frame are related to how the machine is used. Damage from stress cycles is accumulated over time, and measurements throughout the life of the machine are needed to monitor the condition. This results in high demands on the durability of sensors, especially in a harsh mining application. To make a monitoring system cheap and robust, sensors already available on the vehicles are preferred rather than additional strain gauges. The main question in this work is whether the existing on-board sensors can give the required information to estimate stress signals and calculate accumulated damage of the frame. Model complexity requirements and sensors selection are also considered. A final question is whether the accumulated damage can be used for prognostics and to increase reliability. The investigation is performed using a large data set from two vehicles operating in real mine applications. Coherence analysis, ARX-models, and rain flow counting are techniques used. The results show that a low number of available on-board sensors like load cells, damper cylinder positions, and angle transducers can give enough information to recreate some of the stress signals measured. The models are also used to show significant differences in usage by different operators, and its effect on the accumulated damage.

Situation awareness is a crucial capability of any autonomous system, including mining vehicles such as drill rigs and mine trucks. Typically situation awareness is interpreted as the capability of an autonomous system to interpret its surroundings and the intentions of other agents. The internal system awareness however, is often not receiving the same focus, even though the success of any given mission is completely dependent of the condition of the agents themselves. The internal system awareness in the form of vehicle health is the focus of this thesis.

As the mining industry becomes increasingly automated, and vehicles become increasingly advanced, the need for condition monitoring and prognostics will continue to rise. This thesis explores data-driven methods that estimate the health of mining vehicles to accommodate those needs. We do so by utilizing available sensor signals, common on a large amount of mining vehicles, to make assessments of the current vehicle condition and tasks. The mining industry is characterized by small series of highly specialized vehicles, which affects the possibility to use more traditional prognostic solutions.

The resulting health information can be used both to aid in tasks such as maintenance planning, but also as an important input to decision making for the planning system, i.e. how to run the vehicle for minimum wear and damage, while maintaining other mission objectives.

The contributions include: a) A method to use operational data to estimate damage on the frame of a mine truck. This is done using system identification to find a model describing stresses in the structure with input from other sensors such as accelerometers, load sensors and pressure sensors. The estimated stress time signal is in turn used to calculate accumulated damage, and is shown to reveal interesting conclusions on driver behavior. b) A method to characterize the different driving tasks by using an accelerometer and a convolutional neural network. We show that the model is capable of classifying the vehicle task correctly in 96 % of the cases. And finally c), a system for underground road monitoring, where a quarter car model and a Kalman filter are used to generate an estimate of the road profile, while positioning the vehicle using inertial measurements and access point signal strength.

The life and condition of a MT65 mine truck frame is to a large extent related to how the machine is used. Damage from different stress cycles in the frame are accumulated over time, and measurements throughout the life of the machine are needed to monitor the condition. This results in high demands on the durability of sensors used. To make a monitoring system cheap and robust enough for a mining application, a small number of robust sensors are preferred rather than a multitude of local sensors such as strain gauges. The main question to be answered is whether a low number of robust on-board sensors can give the required information to recreate stress signals at various locations of the frame. Also the choice of sensors among many different locations and kinds are considered. A final question is whether the data could also be used to estimate road condition. By using accelerometer, gyroscope and strain gauge data from field tests of an Atlas Copco MT65 mine truck, coherence and Lasso-regression were evaluated as means to select which signals to use. ARX-models for stress estimation were created using the same data. By simulating stress signals using the models, rain flow counting and damage accumulation calculations were performed. The results showed that a low number of on-board sensors like accelerometers and gyroscopes could give enough information to recreate some of the stress signals measured. Together with a linear model, the estimated stress was accurate enough to evaluate the accumulated fatigue damage in a mining truck. The accumulated damage was also used to estimate the condition of the road on which the truck was traveling. To make a useful road monitoring system some more work is required, in particular regarding how vehicle speed influences damage accumulation.

In this work Time Series Classi cation techniques are investigated, and especially their applicability in applications where there are signi cant di erences between the individuals where data is collected, and the individuals where the classi cation is evaluated. Classi cation methods are applied to a fault classi cation case, where a key assumption is that data from a fault free reference case for each speci c individual is available. For the investigated application, wave propagation cause almost chaotic changes of a measured pressure signal, and physical modeling is di cult. Direct application of One-Nearest-Neighbor Dynamic Time Warping, a common technique for this kind of problem, and other machine learning techniques are shown to fail for this case and new methods to improve the situation are presented. By using relative features describing the di erence from the reference case rather than the absolute time series, improvements are made compared to state-of-the-art time series classi cation algorithms.