Impact Cases on AI on construction
If the answer is 42, then what is the question? Perhaps it is the number of challenges faced when implementing AI into Trackunit’s development processes?
There are many high-level discussions about AI solutions taking place right now. Trackunit has undertaken a number of projects this year including a. Machine idle prediction – in production, b. Machine demand forecast – in pilot phase, and c. Meta data classification – in development.
- Machine Idle Prediction: Calculation is done by monitoring engine speed from the CAMbus. Deviations from prescribed ‘normal’ are then notified to the operator. So, the machine is helping the human to be more effective.
- Idle is a problem – 36% of a machines operating lifetime is engine idling. AI pattern recognition used across a wide range of machine engine data to compute specific machine idling.
- Results: decreased service costs, higher resale value, decreased OPEX.
- Machine Demand Forecast: High rental rates make availability a key differentiator for equipment rental. Improving rental availability is highly profitable
- Known – location of rental depots / machine movements (telematics)
- Develop a model to test the hypothesis – based on machine category, weekday and depot, will a machine be rented out?
- Results: Model performance is around 85%. Better customer experience, higher CAPEX utilisation.
- Meta Data Classification: 9 out of 10 connected machines are not properly classified with meta data. Unstructured and disparate data sources make the digital journey time consuming.
- Create an AI model using an NLP model to categorise and structure meta data.
- Worked with partner to add up to 100 data points per machine.
- Result: Awesome fleet management system. Improved customer experience, faster service, improved familiarisation and reduced costs, speeds digital transformation.
Brain of Construction
What if AI didn’t require more than one data scientist?
Democratising AI –Trackunit is introducing the Brain of Construction as a concept.
- Domain knowledge – combined from many available data sources.
- Cluster of models – sharing knowledge from all appropriate models without sharing data.
- Result: Improve site, increase machine value, eliminate downtime.
How to be successful with AI?
Focus on outcomes not outputs. How can we achieve result A, with the available data? Don’t over-elaborate the process.
- Learn and optimise – use ML algorithms, DL algorithms and AI.
- Transform and Aggregate – Clean and detect anomalies, analytics, and metrics.
- Move and Store – Capture as much data as possible.
- Connect and Collect – Instrumentation, logging, sensors, external data, user generated content.