Marine mining initiatives open a new field of subsea operations. Offshore oil and gas sites are still located primarily in areas where divers can support maintenance and repair requirements, but future marine mining will take place in greater depths and with a complexity of machines that requires support from robotic systems equipped with a substantial amount of artificial intelligence (AI). Technologies are being developed that have the potential to support marine mining in all stages from prospection to decommissioning. These developments will likely have substantial influence in the oil and gas industry, itself searching for ways to maximize exploitation of assets.
Robotic Systems Under Current Development
Increasing Autonomous Underwater Vehicle (AUV) Intelligence. Commercial off-the-shelf AUVs rely mostly on acoustic and inertial sensors for their navigation. Speed measurements from a Doppler velocity log are combined with orientation values from gyroscopes and accelerometers to estimate current position. These updates are sometimes augmented by absolute-position fixes from an ultrashort baseline system. However, during such a mission, the inspection assets might not be located exactly at their expected positions. This might be because of incorrect positioning during installation, objects being dragged off location by fishermen, or sediments hiding a pipeline gradually from the view of standard sensors. Therefore, equipping modern AUVs with sensors and software that can search for, detect, track, and reacquire inspection targets is essential.
In addition, classical sensor suites consisting of cameras and sonars can be augmented with higher-resolution 3D sensing such as laser-line projectors (structured light). This enables an AUV’s onboard software to create a millimeter-precision 3D model of the asset, which can be compared with computer-aided-design models or previous-inspection-run data. By using a fully automated 3D-model cross-check, the AUV could detect asset deformations, defects, or marine growth, even while still submerged during the inspection run.
Seafloor AUV Support Infrastructure. Current AUVs have limited endurance, mostly because of limited battery capacity. Depending on the sensor suite, onboard data-storage space also can be a limiting factor. This causes AUV missions to run no longer than a few days at most, depending on AUV size and shape, propulsion, sensor efficiency, and environmental conditions in the deployment area.
To remedy this, current AUV research focuses on subsea docking stations featuring underwater battery charging as well as broadband data links. Depending on the desired charging time, either inductive energy transfer or underwater-pluggable connectors are used. While both require precise positioning, the latter allows for higher charging currents but also requires a more-complex and higher-force plugging mechanism. To transfer inspection results from the AUV and upload new missions, a high-bandwidth data link is required. To charge its battery and transfer data, the AUV has to find its way back home and correctly execute the docking process autonomously. In this application, a combination of dock-relative positioning sensors are used.
The full chain of autonomous dock-relative navigation, docking, undocking, establishing a broadband data link, and recharging the AUV battery while submerged has been developed and successfully tested in the FlatFish project, a joint venture aimed at designing an AUV for conducting repeated inspections of oil and gas subsea structures while being submerged for extended periods of time.
In-System Inspection. Currently, mining machines must be brought back to the surface to be serviced or inspected. This usually interrupts mining operations and incurs high costs. To minimize cost and human intervention, fulfilling these tasks autonomously underwater is desirable. Internal components of mining machines are a particular challenge during inspection because they are usually difficult to access. The in-field inspection of parts and components inside mining equipment thus may require the use of miniaturized underwater vehicles that can operate largely autonomously.
Inspection Sensor Suites
For inspection of a marine mining or production site, higher resolution of survey data allows the detection of smaller problems. However, optical sensors become unreliable in turbid waters and depth information is not available by a single-camera system. To address these problems, a structured light projector can be combined with a camera, allowing increased penetration in turbid water and 3D reconstruction of the observed area through intelligent algorithms. The setup of this combined sensor normally consists of a laser mounted at a fixed distance from, and a fixed angle to, a camera. The combination of distance and angle specifies the depth resolution and working distance of the system. When an object is inside the working area, the laser line on the object can be detected in the camera image and the position of the line can be used to calculate the distances on the line. Doing so while moving the system along the object of interest allows the creation of a detailed 3D model. Experiments have shown that this setup is able to scan objects even in turbid water with minimal reduction in precision, whereas a human operator in the same scenario cannot recognize the object. AI algorithms are needed to filter and match vehicle movement and sensor data to minimize distortion in the resulting object.
Robot Autonomy and AI
Despite significant recent advances in autonomous-vehicle technology, most production scenarios still use systems in which every degree of freedom or feature of a machine has to be controlled directly by a trained expert. Not only to lower costs in human resources, but also to have the ability to use such systems at all—and, furthermore, to use them in a sustainable and secure manner—the level of autonomy must be increased significantly.
These systems usually rely on models. Such models are an abstraction of the reality, allowing the robot to interpret incoming data. For example, a beam-based proximity model of its sonar range finder would allow the robot to interpret the sensor data and deduct physical properties of its environment.
In robotics, these models are mostly analytical models based in physics. Such models require a deep knowledge of the underlying physics but are usually quite computationally efficient once correct parameters are found. However, the models have two main drawbacks. First, the parameter identification and optimization can be very difficult and tedious and often must be reiterated to adapt to changing environments. Second, these analytical models may not be able to describe highly complex or intertwined observations precisely.
In these complex cases, which are quite typical for robotic applications, machine-learning techniques such as neural networks or support-vector machines are AI tools well-suited to develop high-quality models. Admittedly, enough data are needed to train such function regressors as a multilayer perceptron or a support-vector regressor and care must be taken to avoid overfitting.
A severe limitation of currently deployed autonomous systems (e.g., AUVs and autonomous benthic crawlers) is the necessity of being manually retrieved, recharged, and reprogrammed after executing a single mission. For a persistent-autonomy scenario, all of these tasks will have to be executed autonomously. With current technology, such a goal would require a docking station. When a problem occurs that may hinder or prevent the autonomous system from returning to the docking station, the device must decide on its own how to handle this situation without the possibility of human intervention.
These long-term autonomous systems most likely will consist of more than one vehicle. The additional requirement of coordinating more than one vehicle is a demanding one, but breaking down the necessary tasks can be a key to enabling systems to perform them.
Process optimization for future mining activities cannot be achieved by the enhancement of single machines or assets alone, though. Another key component in handling the full chain of future mining processes will be the ability to manage the coordinated operation of multiple complex systems in space and time. Here, manual planning will reach its limits, but AI techniques such as autonomous spatiotemporal planning can be applied to achieve coordination of all actors for productive and safe long-term operation.
石油圈
