Automating a longwall

ACARP’S longwall automation project continues to progress with the main research providers now selected. Prof Paul Lever, chief operating officer, Cooperative Research Centre for Mining Technology and Equipment (CMTE), spoke to ILN about the four areas into which CMTE hopes to make specific inputs.

Staff Reporter

As longwalls have become bigger and more powerful, so too has the imperative increased to remove operators from the harshness of the underground environment, and particularly from proximity to where coal is being cut. Last year the Australian Coal Association Research Program (ACARP) targeted automation of the longwall system as a potential ACARP Landmark Project. See the related story. The aim of the three year project is to deliver significant changes to the way walls are run to as many operations as possible.

The two main research providers selected by ACARP to develop the project are CSIRO’s coal research division and the Cooperative Research Centre for Mining Technology and Equipment (CMTE), both based in Brisbane. Prof Paul Lever, CMTE chief operating officer, spoke to Australia’s Longwalls about the four areas into which CMTE hopes to make specific inputs: knowledge systems, reliability and maintenance, sensors, and implementation and training.

Developing a fully automated longwall system will be a long and complicated process and those involved in the ACARP project know this will not be achieved in three years: “When we’re talking about taking all the people off the longwall completely, and a system able to recognise and react to all ranges of mining conditions, we’re talking ten to twenty years time,” Lever said.

There are simply so many factors that need to be addressed, one of which is the virtual over-supply of information that a longwall generates. Most longwalls today monitor thousands of sets of data from a range of sensor inputs, which represent conditions the longwall is encountering. Many sensors are however, notoriously unreliable, do not work well in all conditions, and present piecemeal data. As Lever points out: “You’re dealing with a knowledge space that’s very fractured and fragmented.”

The issue is further complicated when multiple sensors deliver information about the same event from differing points of view that may conflict. The challenge will be gathering the information generated by the many thousands of data points currently monitored (as well as some additional ones required for automation purposes such as lasers), and converting that information into something useable.

An important challenge is clearly making equipment, particularly sensors, more reliable. Lever said a critical component analysis would be conducted to assess which components generated the biggest risk to automated systems. This risk analysis will highlight components with the highest propensity to fail in automated mode, Lever said. Both CMTE and CSIRO acknowledge the crucial role of original equipment manufacturers (OEMs), without whose commitment the project is probably doomed to fail.

Moving toward full automation will be a staged process, with increasingly complex reasoning processes gradually taken over by computers. The first level of automation will see the converting of data into useful information to allow an off-face operator to monitor the automated longwall operations: “We want the operator to be able to monitor if the longwall components under computer control are operating within its defined bounds and when it is starting to deviate from those bounds,” Prof Lever said.

The next level will be the development of algorithms or systems that will detect when that deviation starts to happen. The system would then alert an operator that the longwall cannot run in automated mode within the required predefined operating bounds. This exception could be the result of an unexpected mining process event, a rapid change in geology or even a possible component failure. At this stage the operator would still make the decision about remedial action as well as its implementation.

As the system became increasingly complex it would eventually deliver both an analysis of what could have gone wrong as well as a suggested action. An operator would still make the final decision about what action to take.

In the final phase of automation, a system would detect a problem, analyse the situation, then develop and implement the solution. Providing a system with the ability to work with numerous data-sets and develop solutions is some challenge, given the complex operating environment of an underground longwall mine.

Within its three-year cycle, the ACARP project could probably deliver elements of the first two levels of automation.

Several approaches can be used for developing higher level, reasoning systems in the dynamic and uncertain longwall environment. Neural networks provide a "black box" system with no ingrained knowledge initially, that delivers certain outputs based on what it has learnt from previous inputs. Lever said OEMs on other automation projects had steered away from using neural nets because of legal implications. “You can’t guarantee that specified inputs will result in the correct associated outputs,” he said.

Alternatively, “fuzzy logic-based control systems are capable of making decisions with information that’s not black or white. Fuzzy systems allow representation of knowledge in semantic form that can use probabilistic associations between data inputs and output actions – closer to the way humans reason.”

Simply put, a fuzzy logic control system can operate like this: the system contains knowledge about how to operate in certain well-defined mining situations. When presented with a new and previously unseen situation that looks partially like one known data set and partially like another, the system incorporates the various solutions from each situation set and merges them to arrive at a wholly new, original solution.

So what exactly can operators expect from the first three years of the automation project?

Lever argues that many deviations in the operation of longwalls are related to a lack of information and/or misinterpretation of information. If an automated cutting system knows what geology lies ahead of the face, and what roof conditions exist, it will be able to cut coal consistently without the wide variations that can occur with manual systems. An operator who does not have information about conditions ahead and has no knowledge of geology can easily lose horizon control which can in turn generate a host of other problems.

Consistency in cutting, which will result in fewer ancillary problems, as well as reduced equipment maintenance, are expected to be two major deliverables within the project’s first three years. The system may be capable of alerting operators to deviation from defined operating boundaries, such as the roof going bad or a chock shutdown.

CMTE also hopes to have input into the development of radar or laser techniques to monitor events such as lumps on the AFC and unusual events coming off the face, which are normally observed by operators on the face.

A final and crucial area will be implementation and training to run automated systems. Running a longwall in automated mode will require different operator skills, with a willingness to allow the equipment to run in automatic mode without interference, Lever said.

Lever, who has been involved with wheel loader automation, said over time the consistent performance of automated systems, which do not push components to their limits and do not get tired, have the potential to deliver better productivity.