Mining is a good example of a problem where no single data stream alone suffices.
A mine might produce telemetry of equipment, location data of machines and people, production results of a process, maintenance history, material-flow events, worker-related signals, among others.
The issue is that those are describing elements of one and the same physical event.
The equipment system reports on the condition of the machine. The location system, of a worker, truck or machine. A production system reports on the events of a process. The maintenance platform has information about its history.
But the question the operator typically asks is "What is occurring across the system, and why does it continue to occur". This is where AIoT becomes of interest.
The physical layer comes from the IoT (Internet of Things)
IoT provides a means to render real world activity as data; in mining this might include machine state, location signals (whether of equipment, of workers, of processes or simply of the material), production measurements or events, and other sensors or machine-related signals.
While many of these signals are relevant alone, they must be stitched together to build a complete picture of the operational context. Consider:
10:32 – Truck A stopped.
By itself this is, almost useless information, additional signals to:
10:32 – Truck A stopped
10:33 – Loading equipment is unavailable
10:35 – material transport from truck A is delayed
10:38 – Truck A has resumed movement
Individual messages now form a temporal sequence and the relevant question becomes one of context and relationship - do the pieces form one picture.
From events to context: Beyond simple IoT sensors
A practical architecture for AIoT therefore takes into account that more than just sensor collection is necessary. For a contextual model for entities, events, processes, locations and their relationships to emerge, a data foundation like the following is required (this is conceptually; exact implementation will vary from operation to operation).
People
Equipment >Location / Identity
Materials
Processes
Operational Events
Event relationships
AI / Analytics
This principle that events should be modeled together as opposed to in isolation applies to any operational domain where assets might be both mobile and interacting with multiple processes.
Why Equipment Telemetry Alone isn't enough.
An AI system working with equipment telemetry detects an abnormal wait period on an excavator. However there can be numerous reasons for this abnormal wait; planned stops, waiting for materials, processing delays, proximity issues with other machines, workload tasks that are not readily observable to simple measurement of usage and/or a workflow event associated with the operator.
The telemetry reports an anomaly; deeper analysis relies upon context beyond just simple measurements. When designing an industrial AI system it is important to distinguish between identifying an event and investigation of the underlying causes, which is required when the data does not explain the surrounding context.
The advantage of using location is an operation signaling mechanism.
Location as an operating signal is particularly beneficial in mining because a large number of actors and machines involved in the mining process are mobile.
An interesting relationship is:
Worker
Machine
Material
Process
Any of those could be associated with a particular physical location. For instance if two excavators stop repeatedly interacting with a particular conveyor it may be worth investigating whether that conveyor stage can be part of the mining operational analysis. Its useful to identify not the events of individual machines, but interactions within an entire operational context-such as those made discoverable by common locationing signals.
The advantage this provides to the operator is that an engineer may need to understand this relation as one of two pieces in an investigative task or may need to visualize it as one component in a workflow.
Location can sometimes itself become an operational signal rather than just another layer to visualize what happens and where.
Therole of the engineer/AI system: from question to hypothesis.
AI itself should not just be dropped on top of any available dataset. Rather, an engineer should approach problem solving in a workflow as follows:
What is the question?
Identify the involved entities and the physical process steps: What should happen? Identify relevant signals and events; what can I learn from here that makes sense for operation?
Identify the data sources; how will I fill the knowledge gaps?
Build a context for event relationships, discover anomalous patterns. Use these derived relationships to answer operational questions. Ultimately, the delivered analytics should lead to a decision that an operator can action. An example question isn't \"where can we apply AI\", but "Why does my equipment constantly halt in one phase?".
Potential solutions would include analysis of utilization of assets, troubleshooting operational problems. Mining applications of AIoT are described with a lot more detail from mining and resource grouping here, covering critical mineral, mineral processing and other mining applications.
Physical Design of Systems
The challenge within industrial contexts when designing an AI system is that each event is born from the physical world, and so can be understood through context, instead of abstract concepts alone. It takes form through a simplified framework like the below:
Prediction / anomoly detection
Equipment waiting-unusual patterns detected.
Associated contextual observation (work cycle) - repeated waits near a transfer points.
Investigation. - inconsistencies in upstream delivery of materials to the processing stages.
Operational question - what is occurring at the transfer stage.
Decision/action- review the interaction at the transfer stage to optimize it.
The output itself is far less significant than the chain of context leading to it. It helps an individual to determine what to do next.
Practical checklist for building AIoT systems in mining:
- What is the desired outcome or operational question? Always build the data set around a key question about the operation.
- What entities are associated with the operational questions (people, processes, objects/machinery, resources etc.)?
- What events signify the operational behavior desired? Identify what 'happenings' constitute the operation (movement, processing, transfers of materials, human intervention etc.).
- How are these events interrelated? Use time, location, identity of assets and individuals involved, or process dependencies.
- What data are needed to answer the question? Ensure there is information on each part of the answer/hypothesis being built-otherwise a model may give a statistically interesting answer that doesn't apply.
- How can this be validated and how will each result be measured? Operations will require operational validation.
- What specific decision or process improvement does this knowledge bring. Unless the analyst's insights allow action, the work loses meaning.
The question not the model: best practice for AIoT design in mining.
As engineer, one important guideline when approaching AIoT projects within a mine is to begin by identifying the underlying problem to be solved and thus form the specific question that guides every element of the analysis rather than "where can we apply AI.". It is likely the engineering will need to be accompanied with substantial rework in connectivity, understanding the data, and identifying the relations between events. Mines are abundant with sources of data, however few operations have explored bringing their operational context into an AI enabled environment for an holistic, intelligent overview; by connecting people, machines and processes together. AIoT can provide context; the essential component needed from sensing data and AI itself, to turn detection into understanding of what's happening. For More Info Visit: apertureventurestudio.com
Top comments (0)