AI: How telcos can profit from deep learning

(Re)Connecting with Consumers

Login to access

Want to subscribe?

This article is part of: (Re)Connecting with Consumers

To find out more about how to join or access this report please contact us

Artificial intelligence (AI) is improving rapidly thanks to the growing use of deep neural networks to teach computers how to interpret the real world (deep learning). These networks use vast amounts of detailed data to enable machines to learn. What are the potential benefits for telcos, and what do they need to do to make this happen?

The enduring value of connected assets

In the digital economy, the old adage knowledge is power applies as much as ever. The ongoing advances in computing science mean that knowledge (in the form of insights gleaned from large volumes of detailed data) can increasingly be used to perform predictive analytics, enabling new services and cutting costs. At the same time, the widespread deployment of connected devices, appliances, machines and vehicles (the Internet of Things) now means enterprises can get their hands on granular real-time data, giving them a comprehensive and detailed picture of what is happening now and what is likely to happen next.

A handful of companies already have a very detailed picture of their markets thanks to far-sighted decisions to add connectivity to the products they sell. Komatsu, for example, uses its Komtrax system to track the activities of almost 430,000 bulldozers, dump-trucks and forklifts belonging to its customers. The Japan-based company has integrated monitoring technologies and connectivity into its construction and mining equipment since the late 1990s. Komatsu says the Komtrax system is standard equipment on “most Komatsu Tier-3 Construction machines” and on most small utility machines and backhoes.

Komatsu’s machines ship with GPS chips that can pinpoint their position, together with a unit that gathers engine data. They can then transmit the resulting data to a communication satellite, which relays that information to the Komtrax data centre.

The data captured by Komtrax (and other Internet of Things solutions) has value on multiple different levels:

  • It provides Komatsu with market intelligence
  • It enables Komatsu to offer value added services for customers
  • It gives detailed data on the global economy that can be used for computer modelling and to support the development of artificial intelligence

Market intelligence for Komatsu

For Komatsu, Komtrax provides valuable information about how its customers use its equipment, which can then be used to refine its R&D activities. Usage data can also help sales teams figure out which customers may need to upgrade or replace their equipment and when.

Komatsu’s sales and finance departments use the findings, for example, to offer trade-ins and sales of lighter machines where heavy ones are underused. Its leasing firm can also use the information to help find customers for its rental fleet.

Furthermore, Komatsu is linking market information directly with its production plants through Komtrax (see Figure 1). It says its factories “aggressively monitor and analyse the conditions of machine operation and abrasion of components” to enable Komatsu and its distributors to improve operations by better predicting the lifetime of parts and the best time for overhauls.

Figure 1: How Komatsu uses data captured by its customers’ equipment

Source: Komatsu slide adapted by STL Partners

Value added services for customers

The Komtrax system can also flag up useful information for Komatsu’s customers. Komatsu enables its customers to access the information captured by their machines’ onboard units, via an Internet connection to the Komtrax data centre.

Customers can use this data to monitor how their machines are being used by their employees. For example, it can show how long individual machines are sitting idle and how much fuel they are using. Komatsu Australia, for example, says Komtrax enables its customers to track a wide range of performance indicators, including:

  • Location
  • Operation map (times of day the engine was on/off)
  • Actual fuel consumptionAverage hourly fuel consumption
  • Residual fuel level
  • High water temperature during the day’s operation
  • Dashboard cautions
  • Maintenance reminders/notifications
  • “Night Time” lock
  • Calendar lock
  • Out of Area alerts
  • Movement generated position reports
  • Actual working hours (engine on time less idle time)
  • Operation hours in each work mode (economy, power, breaker, lifting)
  • Digging hours
  • Hoisting hours
  • Travel hours
  • Hydraulic relief hours
  • Eco-mode usage hours
  • Load frequency (hours spent in four different load levels determined by pump pressures or engine torque)

 

Content:

  • Introduction
  • Executive Summary
  • The enduring value of connected assets
  • Tapping telecoms networks
  • Enabling Deep Neural Networks
  • Real world data: the raw material
  • Learning from Tesla
  • The role of telcos
  • Conclusions and Recommendations

Figures:

  • Figure 1: How Komatsu uses data captured by its customers’ equipment
  • Figure 2: Interest in deep learning has risen rapidly in the past two years
  • Figure 3: Deep learning buzz has helped drive up Nvidia’s share price
  • Figure 4: The key players in the development of deep learning technology
  • Figure 5: Mainstream enterprises are exploring deep learning
  • Figure 6: The automotive sector is embracing Nvidia’s artificial intelligence
  • Figure 7: Google Photos learns when users correct mistakes
  • Figure 8: Tesla’s Autopilot system uses models to make decisions
  • Figure 9: Tesla is collecting very detailed data on how to drive the world’s roads