adaro mining

Real-time Water Level Monitoring & forecasting

Using continual deep learning models on historical data & real-time water level sensors to elevate Adaro Mining’s logistical operations

Challenges

A leading coal operator in the region, Adaro rely on river transportation for their coal mining operations in the Kalimantan province of Indonesia. Its fleet of 40 to 45 barges ferry the coal through the 1,090km long Barito river arriving at the destination port for the seaborne coal trade.

The hourly and seasonal fluctuations can see certain segments of the river completely dry up, resulting in barges being stranded, damaged, and incur a safety risk to its operators. When water level is too high, barges sailing through certain segments of the river are at risk of collision with the canal bridges at different junctures of the Barito river.

Without the combination of data and predictive analytics, operators have limited means of knowing in advance the water level at different points of the river, and sailing on water level that is either too low or too high can have expensive and devastating consequences, from a personnel safety perspective in addition to causing operational and logistical disruption, both of which are extremely costly.

Operations and Logistics Planning department have difficulty in ensuring the right amount of barges are provisioned beforehand if there is no way of forecasting the number of sailable days. Teams on the ground have to rely on gut instincts in deciding on an appropriate load capacity taking into consideration the fluctuating water level along Barito river.

At a glance

Challenges

Benefits

$ 700
labor cost reduction
700
man-hours saved every month
~$ 1000000
labor cost reduction

Approach

The transition from manually-kept spreadsheet data, to a cloud-native, IOT-enabled analytics solution for Indonesia's second largest producer of thermal coal
  • Cloud First and Digitalization

    Adaro Mining Technology (AMT), led by Pak Eri Basuki, provided Supertype with up to 13 years of historical water level at key sections of the Barito river. The data are collected manually in spreadsheets. A data pipeline is constructed by our engineers to extract, transform (clean) and load the data into a centralized BigQuery data warehouse.

  • Queryable and Synchronization

    AMT and Supertype work together on the necessary architectural design between how data could be queried and consumed by the various supporting applications within the analytics stack. A queryable interface is created by our engineers to allow for data retrieval, updates and consolidation in real-time.

  • Predictive Analytics and Continual (Deep) Learning

    We develop a highly tuned model using the latest in neural networks research to predict water level for the days and weeks ahead. In later phases of this project, AMT installed water level sensors for the continuous water measurements of the water level at different spots of Barito river. Our cloud infrastructure along with the data pipelines made it easy to stream these measurements directly to our cloud storage in 10-minute intervals, and the forecast model is continually re-trained on Google's Vertex AI platform to produce up-to-date forecasts using incoming data.

"The Supertype team has demonstrated high capability in understanding our challenges, which were relatively new to them in a relatively short time. The resulting model also has a high degree of accuracy, which we can use to reduce the element of surprise in our dynamic coal supply chain. Currently, it's like having a special ability to predict the very dynamic of natural conditions. We are very satisfied with the work of Supertype."
eri basuki adaro
Eri Basuki
President Director, Adaro Mining Technologies

Qualifying the value of Supertype’s data science & engineering work

Deliverables & Key Results

Key Technology
key technology used by supertype team

Robust Data Warehousing Infrastructure

A proper, well thought-out cloud architecture that uses proven, modern data warehousing technology that consolidates:

  1. all 13 years of historical water level measurement
  2. incoming data streams from multiple water level sensors installed at various points along the river
  3. prediction of water level that is the result of a continually trained deep learning model
  4. other schematic information
Our engineers also developed a suite of REST API service to enable ad hoc querying as well as CRUD access to the centralized data warehouse

Automatic and Scheduled Predictions

  • Custom-trained deep learning models that are continually trained and deployed on Vertex AI, to return accurate prediction of water level at key checkpoints of the Barito river with upwards of 91% accuracy
  • A fully automatic machine learning task that generate predictions for water level daily; as well as a 3-month forecast for the number of loadable days (to be distinguished between half-load, full-load)

Accessible Analytics Interfaces

  • For the business user, Supertype's business intelligence developers create a Google Data Studio (Looker Studio) dashboard as a business support tool, aiding in decision-making through visualization, analytical charts, first- and third-party data (such as weather condition in the Barito river basin
  • A slick, integrated web application developed with React and modern server-side technologies to help the client acquire key insights on water level, query data, inspect water level forecasts, perform two-way synchronization with the data warehouse and access business intelligence features

Benefits

Automatic and Scheduled Predictions

The periodical 3-month forecast generated by Supertype produced an MAPE of ~9%, and is critical at helping Adaro procure just the right amount of transportation barges for the months ahead, producing up to $1,890,000 in savings annually from un-utilized barges.

Automatic and Scheduled Predictions

The EL (Extract - Load) pipeline provides a real-time data streaming mechanism for payloads of sensor data from the installed sensors. These data packets are streamed and immediately buffered into BigQuery on a 10-minute interval, at a total transfer rate of ~6s, making it available for ad hoc query, exploratory analysis and machine learning procedures.

Along with the automated machine learning inference and retraining pipeline, the data warehousing and automated pipeline save an estimated 920 man-hours every month (~$150,000/year) in labor cost.

Contact Us

Want more case study?

This case study, along with others in our archive is also available as a PDF. To schedule a call with us and learn more about our past projects, leave your contact method below.

Your work email, so we can reach you. We respect your privacy and will not add you to any mailing list.

Inquiries

s@supertype.ai

More content from Supertype

×

Table of Contents