Open Cut » Overburden Removal
The objectives of the project were to:
- Demonstrate the feasibility of automatic sensing of bucket position.
- Characterise the electro-dynamic dragline system.
- Demonstrate closed-loop operation of the ADPC tenth scale dragline rig using bucket position feedback.
Bucket position sensing
The major sensing requirement for closed-loop control of bucket position during swinging and dumping is to develop a system to measure the position of the bucket in space. This is the component missing from previous work.
Demonstrate closed-loop control
To show the feasibility of automatic bucket positioning it is necessary to demonstrate both the operation of a bucket position sensor and the integration of this sensor input into a control system controlling the machine hoist drage and swing drives.
Results
Bucket location
A successful bucket location system was developed using machine vision. Because the pendulum swing of the bucket below the boom is the only variable not measured by existing monitoring systems, the machine vision system is required to measure only the pendulum motion. This was achieved using a video camera mounted at the boom point looking downwards towards the area in which bucket motion occurs.. The machine vision system was implemented on a full-sized machine at Tarong and on the ACIRL ADPC scale dragline rig at Riverview.
Techniques to identify the bucket against the background based on contrast thresholding and image edge detection are described. The compotation effort required to identify the bucket is reduced by the use of a processing rectangle which uses other information about bucket position to decrease the amount of the camera field of view that must be processed in real time.
Feasibility of dragline control
Automatic control of the ADPC scale-model dragline was achieved. It was necessary to use a target on the hoist rope to ensure reliable operation of the machine the bucket, after an abrupt halt of swing motion, by the use of the machine vision was demonstrated and quantitative results are presented.
Conclusions
Closed-loop control
Based on the successful experimental program, closed-loop control of a full scale dragline in the swing-dump-return motion is feasible. In fact the control problem to automate a full seize machine is simpler than the one already solved to automate the model in this project.
There are two main reasons for this. Full scale machines are better documented in terms of the electrical and mechanical characteristics of the machine. Secondly, on full scale machines, time constants are approximately three times longer, making computation easier.
Closed-loop control of a dragline was demonstrated convincingly on the ADPC rig. In this project, actual bucket position was controlled for the first time using a machine vision based control system. All the techniques used, including characterisation of the electrical and mechanical systems are able to be transferred directly to a full sized machine. In fact, from a control theory viewpoint, the problem becomes simpler in a full sized machine because electrical and mechanical time constants are a factor of approximately three longer, reducing the sampling demands on the sensing and control system.
Bucket position sensing
The 'best' method for bucket position sensing has not been found. Identification of bucket position itself is subject to problems from dust and lighting conditions, and the use of a target on a full sized machine would have to be a last resort. Promising results have been obtained using a system which measures the angle of the hoist ropes themselves close to the boom point sheaves.
Automatic sensing of bucket position to provide performance adequate for closed-loop control of the ADPC rig was successfully demonstrated with a machine vision system using a target attached to the hoist rope. This system worked well in daylight but was not tried at night under lights. While not desirable, this approach could be used in the field on a full sized bucket. The target would need to be of the order of 3 metres in diameter. It would require periodic cleaning, perhaps once per shift.
The conclusion from this experiment and from work on the Tarong machine is that it is unlikely that machine vision, segmenting the basic bucket shape from the background, will provide a robust system. To locate the bucket itself reliably it would be necessary to attach some form of target.
A rope following, machine vision solution was investigated on the Tarong machine. This approach shows real promise. It will be necessary to provide lights on full sized machines to illuminate the area of the hoist ropes just below the boom point sheaves to evaluate this method at night.
Feasibility of implementation
In order to partially automate the ACIRL scale dragline in an experimental fashion, this project has used very sophisticated image processing and control techniques and multiple networked computers to achieve the successful result. This could give the impression that the system required in a full scale machine will also require specialised site knowledge. This is not the case. In the commercial system the cost will be relatively low and the hardware complexity will be of the same order as a Tritronics monitoring system.
Future Work
On the basis of the results of this project, further funding of $817,000 has been secured to demonstrate closed-loop control of a full sized dragline from a syndicate comprising ACARP, BHP Australia Coal, and CRA (coordinated by Pacific Coal). The test site has been chosen as Meandu mine operated by Tarong Coal and the researchers will be the CMTE (CSIRO and the Julius Kruttschnitt Mineral Research Centre) and Tritronics Australia.