by Bernard Marr
As Daniel Jeavons, general manager for data science at Shell, told me: “If you think about it, as a grid operator you're operating many, many electric charging posts … if all the cars all plug in at the same time and automatically start charging, you create a big load on the grid … which, by the way, can't be filled by solar, because it's 7 am or 8 am in the morning.”
This means that the grid supplying the charging stations will have no choice but to use energy generated from fossil fuels to provide your electricity, somewhat negating the good you are doing the planet by driving electric in the first place!
How Is Artificial Intelligence Used In Practice?
Shell's system, which it leases to electric vehicle charging site owners, uses AI to power analytics, which can spread the load caused by rush-hour demand for charging.
By getting to understand the patterns in the peaks and troughs of customers using the charge points, it builds up a profile that can be used to predict energy demand.
This effectively allows the charging point network to stagger the use of energy through the day, ensuring power is always available when it is needed, and it isn't wastefully generated at times when it isn't.
Jeavons says: “So what we can do by understanding people's charge profiles is we can spread the load during the day, which basically means we can save the consumer money. But also more renewables are used – because if you can charge more people at lunchtime, there's going to be more solar on the grid at that point.
“It's an example of where we see the role of artificial intelligence playing a key part – thinking about not just how we can make things more efficient, but also how we can change energy consumption patterns to take more advantage of renewable sources.”
What Technology, Tools And Data Were Used?
Shell supplies the entire end-to-end network used by the charging points. This includes generating the power itself, installing and monitoring the charging stations, processing data in the cloud, and developing and supporting the app used by drivers to interface with the system.
This means that it is able to collect data on every step of the process. The app drivers use to control the charge also allows Shell to profile drivers to understand how demand changes around the day at its charging points.
Power cost savings generated by the cuts in wastage can be passed back either to the customer by reducing the cost of the charge or to the site owner (depending on the owner's business model).
What Were The Results?
So far, Shell's RechargePlus has been rolled out in California, where results will be monitored before the technology is rolled out in other territories.
Although there are no concrete results published yet, the experience it has given Shell with rolling out AI technology is likely to prove very valuable when it comes to planning further deployments.
Jeavons says: “What it means in practice is that we as a data science team are in a great position, because we can make our current business more effective, more efficient, more reliable, safer – by applying AI into those settings – and that's great.
“But we can also play a role in creating some of the new business models that we want to create and that's really exciting, because we're playing our part in taking Shell into the next generation of energy sources, new fuels, and new sources of revenue.”
Key Challenges, Learning Points And Takeaways
Shell employs AI solutions across its business with key use cases around meeting its energy transition targets.
Drivers often cite a lack of available charging infrastructure as a reason for continuing to choose fossil fuel-powered vehicles.
Charging site owners don't like to front the cost of installing infrastructure before the user base is in place – but supplying infrastructure-as-a-demand helps them to share some of the risk with Shell.
AI can be used to understand and predict energy demand at recharging points, and can regulate supply to avoid adding unnecessary strain at peak times.
Notes
1Fortune, Royal Dutch Shell: http://fortune.com/global500/royal-dutch-shell/
2Autotrader research published in PV Magazine, UK drivers don't plan on buying an electric car for almost a decade: https://www.pv-magazine. com/press-releases/uk-drivers-dont-plan-on-buying-an-electric-car-for- almost-a-decade/
48
Siemens: Using Artificial Intelligence And Analytics To Build The Internet Of Trains
Siemens AG is a German industrial conglomerate that manufactures and sells transport machinery, medical equipment, water treatment systems and alarm systems, as well as provides financial and consulting services.
In recent years, Siemens has been rolling out its vision for what it calls the “internet of trains”. This is the on-rails segment of the wider Internet of Things – the concept that devices of all shapes and sizes can be networked through the cloud and empowered to talk to each other.
With the market for “smart railways” products and services forecast to grow from $11 billion in 2017 to over $27 billion by 2023,1 Siemens is making its play for its share of the cake with its Railigent predictive artificial intelligence (AI) platform.
What Problems Is Artificial Intelligence Helping To Solve?
All over the world, time and money are wasted due to delays occurring on public transport networks. If people or goods aren't in the place they're supposed to be at the time they are needed, then business doesn't get done.
Knock-on effects of this include the fact that people often choose more reliable though far more environmentally damaging alternatives – air travel – over rail travel, when they can't risk running late or missing an appointment.
Railway delays can be caused by inefficient scheduling when projected passenger numbers or time taken between departure and arrival are incorrectly forecast, as well as equipment faults and breakdowns.
How Is Artificial Intelligence Used In Practice?
Sensors and cameras are used to measure how every part of the transport system is moving and operating.
This enables a “digital twin” model of a rail system to be built and used to forecast when factors likely to lead to delays or inefficiencies will emerge, and what can be done to either quickly react or prevent them from occurring in the first place.
The insights serve three primary purposes. First, they can improve asset availability by ensuring both that trains are in the right place at the right time, and that breakdowns and faults can be remedied far more quickly by enabling servicing and repairs to be done more efficiently.
Second, they can optimize energy efficiency across the transport network. This means that energy usage can be measured and predictions made about when and where power will be needed. This can reduce the environmental impact of trains – already considered relatively environmentally friendly – even further.
Energy conservation during train travel can be optimized too – with a better understanding of what is going on at a macro level within a railway network, trains have to brake less frequently, conserving energy needed to push them forwards. They can also more reliably travel at higher speeds, leading to shorter journey times.
Third, asset utilization can also be improved. This means more accurately forecasting of the number of passengers, or the amount of freight, that will be transported between destinations in a given time. The fewer train journeys necessary to move all of the passengers or goods, the lower the environmental impact and financial cost to the operator will be.
What Technology, Tools And Data Were Used?
Siemens calls its connected trains platform Railigent, which in turn connects to Mindsphere, its industrial Internet of Things operating system.2
Sensors aboard the trains capture everything from engine temperature and rail vibration frequency to the open or closed state of doors, and image data from external cameras is collected and processed to identify factors that can cause delay. In one UK pilot project, 300 sensors were used, generating 1 million sensor log readings
over a one-year period.3 Data collected from the sensors is correlated with breakdown and downtime data.
As well as internal data collected from the train itself, external data such as camera feeds is also used. This allows the trains to capture images of the track ahead, allowing for faults to be automatically recognized and the locations where faults will develop in the future to be more accurately predicted.4 It also improves worker safety by reducing the need for humans to make manual inspections on active tracks.
The system is designed so that data can either be transmitted from trains in real time using mobile data networks or, for regions with poor coverage, uploaded when a train arrives at its destination.
Siemens worked with Teradata's Aster discovery platform to pull insights from the data generated by the sensors.5 Data can be relayed to control rooms through a dedicated reporting and visualization platform, or it can be integrated into tools that are already used. Critical reporting and events can also be sent via SMS.6
What Were The Results?
As Gerhard Kress, director of mobility data services at Siemens, puts it: “Our customers get more mileage from fewer trains and, therefore, use their assets better while reducing their costs. Additionally, data analytics can speed up the root-cause analysis, reducing labor time.”
While working with one German rail operator, Siemens managed to predict every single component failure within the bearings, gearboxes, motors and other mechanical elements.7
A critical result is that Siemens now feels so confident in the accuracy of its forecasting that it is able to offer its customers uptime guarantees.
It hopes to improve the efficiency of trains to the point where they can be competitive with airlines, then there will be important environmental gains to be made, too.
Key Challenges, Learning Points And Takeaways
Reducing delays and minimizing environmental impact are the key drivers to the move towards a smart automated system in rail networks.
Sensor data can be overlaid with operational data such as breakdown and maintenance reports to give a fuller understanding of factors that cause delays when it is used to train AI systems.
Increasingly, unstructured data such as visual data from camera feeds will be a key ingredient in this mix. Image recognition software helps make sense of this unstructured data by turning it into information that machines can understand and correlate with other data sources.
Notes
1Gartner, Global Smart Railways Market Research Report – Forecast To 2023: http://garnerinsights.com/Global-Smart-Railways-Market-Research-Report—Forecast-to-2023
2Siemens, MindSphere – The Internet of Things (IoT) Solution: https:// www.siemens.com/global/en/home/products/software/mindsphere.html
3Teradata, The Internet of Trains: http://assets.teradata.com/resource Center/downloads/CaseStudies/EB8903.pdf
4Siemens, Railigent® – the solution to manage assets smarter: https:// www.siemens.com/global/en/home/products/mobility/rail-solutions/ services/digital-services/railigent.html
5Teradata, The Internet of Trains: http://assets.teradata.com/resource Center/downloads/CaseStudies/EB8903.pdf
6Siemens – The Internet of Trains 2.0
7Forbes, How Siemens Is Using Big Data And IoT To Build The Internet Of Trains: https://www.forbes.com/sites/bernardmarr/2017/05/30/how-siemens-is-using-big-data-and-iot-to-build-the-internet-of-trains
49
Tesla: Using Artificial Intelligence To Build Intelligent Cars
Tesla is a pioneer in the development and marketing of electric cars. It also has a large stake in the future of autonomous vehicles – in fact every Tesla ever produced has the potential one day to become self-driving through software upgrades. It also manufactures and sells advanced batteries and solar panels.
Autonomy in cars is graded on a scale from one to five. Features such as adaptive cruise control and automated parking systems are classed as level 1, while fully autonomous vehicles, capable of driving anywhere on- or off-road with no driver intervention, are classed as level 5.
Tesla's founder and CEO, Elon Musk, has said that he believed his company's vehicles will achieve full (level 5) autonomy in 2019.
What Problems Is Artificial Intelligence Helping To Solve?
Driving requires human beings to be consistently at a high level of awareness for long periods of time. As the behavior of other drivers on the road, as well as circumstances such as the weather and road conditions, can be highly erratic and unpredictable, it's no surprise that over 40,000 people were killed in road accidents in the United States alone in 2017.
Of course, minor (not causing fatality or serious injury) accidents happen far more regularly – leading to a huge waste of resources and time.
And even if you don't have an accident, time spent driving is time that could be spent doing other things – whether being productive in a work capacity, spending quality time with co-passengers or friends and family who aren't present, through social media, or just catching up on sleep!
How Is Artificial Intelligence Used In Practice?
When it comes to self-driving vehicles, artificial intelligence (AI) is used to make decisions based on road conditions around the vehicle, such as the direction it is travelling, the planned destination and the behavior of other traffic in the vicinity. Camera data is processed using computer vision technology to allow the car to understand what it is “seeing” and react accordingly.
This operates at three levels – internal (information gathered and processed internally by the car), global (information gathered across an entire fleet of autonomous vehicles and shared between them) and local (information gathered by “ad hoc” networks of autonomous vehicles within close proximity to each other). When autonomous cars are commonplace, this is likely to be supplemented by data from networks formed between other machines – such as traffic cameras, roadside sensors and even pedestrians’ mobile phones.
Tesla's current level 2 autonomous driving system – known as Autopilot – allows the car to match speeds to traffic conditions, change lanes on a motorway, transition from one road to another, self-park and be “summoned” to and from a parking location. However, drivers must remain fully engaged with the vehicle and be ready to take over control at any moment.
AI in cars poses some important ethical questions that have not yet been fully resolved. For example, how should an autonomous car react when given the choice of hitting a small child who has fallen over in the road, and taking evasive action such as driving off a road, potentially harming the driver or other people? A human in the same position would be forced to make the same choice, of course – with no more guarantee that they would make the “correct” decision than a robot would. It could be argued that, given enough data – which unfortunately would require a few “incorrect” decisions before it could be gathered – an automated car would be able to calculate the least catastrophic scenario and take action more reliably than a human.
In the more immediate future, Tesla is believed to be working on a Siri-style AI assistant that can communicate with drivers vocally while we still have to pilot our vehicles ourselves.
Responding to a question on Twitter, Musk stated in early 2018 that soon Tesla drivers would be able to do “pretty much anything” through voice controls. This implies that AI will be used to interpret commands using natural language processing technology for the car to understand what a driver means when they use a particular phrase.
What Are The Results?
Tesla says that its Autopilot system can cut accidents by 40%. This figure has come under scrutiny with some commentators saying there isn't enough data available to prove that it's correct, and that it hasn't been independently verified. In response, Tesla has stated that it will begin reporting its safety and accident data quarterly. So far there have been two fatal accidents involving Teslas using Autopilot, and the US National Highways Transport Safety Administration says there is still “no clear evidence” of an inc
rease in safety.
However, Tesla reports that before its Autopilot system was activated, the rate of airbag deployment was 1.3 per million miles driven. Following its activation, that rate fell to 0.8.
What Technology, Tools And Data Were Used?
Just as information informs and trains human intelligence, data is the lifeblood of AI. Tesla's fleet of electric vehicles is equipped with an extensive array of sensors. These include cameras scanning the road, atmospheric sensors for monitoring weather conditions and even steering wheel sensors to understand how drivers use their hands while operating the vehicles.
All of this data is processed via machine learning algorithms to understand what is relevant to the car's operation, and how best the car should act or react to any given circumstance to safely navigate itself from A to B.
The number of Tesla vehicles already deployed on the roads – and constantly gathering and uploading driving data to the cloud – means that Tesla has a head-start on other car makers vying for the lead in the race to develop autonomous vehicles, which are mostly still using prototypes.
Following a partnership with Nvidia to develop the first generation of intelligent driving software, Tesla has now said that it is working on its own AI algorithms internally.
Key Challenges, Learning Points And Takeaways