by Bernard Marr
At the same time, environmental concerns mean there is an increasing need to find cleaner, safer and less polluting sources of energy. Wind, solar and tidal energy are expected to play a bigger role in filling our energy needs as time goes on.
Currently, power is often wasted because it is generated inefficiently or because demand has been inaccurately forecast. Increasing or decreasing the power output of a station is a costly process, and factors such as changing weather conditions can cause unexpected surges or slumps in demand.
How Is Artificial Intelligence Used In Practice?
GE is working towards developing what it calls the “digital power plant”, which it believes will be the first step towards creating a global “internet of energy”.
At GE Power, engineers have leveraged big data, machine learning and predictive analytics to better understand the stresses and demands at work in a modern power station. Implementation of the system allowed one power plant at Chivasso in Italy, which had previously been taken offline due to its inability to respond quickly enough to changing demands, to be brought back online with half the environmental footprint it previously had.3
Sensor data from machinery throughout a plant is analyzed by machine learning algorithms, which are able to identify optimum operating parameters or highlight issues that may be causing previously unnoticed inefficiency.
In practice, this means that production can be increased or decreased when changes in demand are anticipated, and faults can be remedied before they become major problems, through predictive maintenance.
What Technology, Tools And Data Were Used?
GE's “internet of energy” is built around their industrial internet platform Predix. It enables GE to take a global overview of energy production at its customers’ global network of plants, which include everything from coal, gas and nuclear to wind and solar farms.
A typical plant is equipped with over 10,000 sensors, which monitor every aspect of operation, each generating around 2 TB of data per day.4 The Predix platform is designed to be able to read sensor data from all machinery in a plant, not just machines manufactured and sold by GE.
GE used this data to pioneer the concept of the “digital twin” – which is a computer simulated copy of any part of their business, accurately showing how it is impacted by real world factors such as increases in demand and changeable weather.
This means that power station operators, such as Exelon, which has installed the Predix system across its US network of power stations,5 can more accurately forecast factors that might affect operating conditions. For example, by more accurately forecasting the weather, it is able to understand when solar farms will be least effective, and there may be a need to ramp up output at gas-fired plants.
GE Power categorizes these aspects of its “internet of energy” program as “asset performance management”.
In addition, artificial intelligence (AI) is utilized in business optimization – using software from Tamr6 it has applied machine learning to managing its enormous procurement operations. The many divisions of GE purchase hundreds of thousands of items of inventory from a global network of suppliers, previously with little central coordination. By training the system on invoices and purchase records, they are able to avoid over-ordering and ensure cost-efficiency in an environment where multiple departments may all be procuring the same items from different suppliers.
What Were The Results?
GE Power's chief digital officer, Ganesh Bell, told me: “We have seen results like reducing unplanned downtime by 5%, reducing false positives by 75%, reducing operations and maintenance costs by 25% – and these start adding up to meaningful value.”
Using the Tamr platform to manage procurement and inventory processes led to a saving of $80 million over three years, according to vice president of technical product management for GE Digital Thread, Emily Galt.
Key Challenges, Learning Points And Takeaways
The power industry needs to increase its output by around 50% over the next 20 years, while simultaneously cutting its carbon footprint by 50%.7 Advanced analytics powered by AI has the potential to help with this mission.
More accurately predicting peaks and troughs in the demand for energy within a geographic region means increased efficiency and less waste.
Today, just about any machinery can be connected to the cloud and start to generate data. But the real value lies in learning how to interpret that data and draw insights from it. With hugely complex data such as machine logs generated by power station equipment, AI is uniquely capable of doing this.
Notes
1GE, GE Reports: https://www.ge.com/reports/energy/
2GE, Waking Up as a Software and Analytics Company: https://www.ge .com/digital/blog/waking-up-software-analytics-company-building- intelligence-machines-systems
3GE, Breathing new life into old assets: https://www.ge.com/power/case-studies/chivasso#
4Fool.com, 3 Ways General Electric and Exelon Are Cashing in on Digital: https://www.fool.com/investing/2016/12/22/2-ways-ge-is-making- digital-indispensible.aspx
5GE, The Internet Of Electricity: GE And Exelon Are Crunching Data Generated By Power Plants: https://www.ge.com/digital/blog/internet-electricity-ge-and-exelon-are-crunching-data-generated-power-plants
6Fortune, GE Saved Millions by Using This Data Startup's Software: http:// fortune.com/2017/05/17/startup-saved-ge-millions/?iid=sr-link2&utm_ campaign=GE%20Saves%20Millions
7Fool.com, 3 Ways General Electric and Exelon Are Cashing in on Digital: https://www.fool.com/investing/2016/12/22/2-ways-ge-is-making-digital-indispensible.aspx
43
John Deere: Using Artificial Intelligence To Reduce Pesticide Pollution In Agriculture
John Deere was founded by a small-town blacksmith as a toolmaker, and over 150 years later has become one of the world's leading manufacturers and suppliers of agricultural and industrial machinery.
It has always been a technological innovator – investing in gasoline engines to mechanize its farming machinery in the early 20th century, and in GPS technology to begin the march towards automation in the late 1990s.1
Over the last decade, John Deere has transformed into more of a technology company – selling data as a service to allow farmers to make better informed decisions when it comes to running their operations. In addition, the company is offering autonomously driving tractors,2 intelligent sensors and software and even agriculture drones.3
What Problem Is Artificial Intelligence Helping To Solve?
The world's population currently sits at around 7.5 billion and is expected to grow to over 9 billion by 2050.4 Feeding all these hungry mouths is going to require increasing the amount of food we produce by 70%, according to the United Nations Food and Agriculture Organization. At the same time, due to increasing urbanization, climate change and soil degradation, the amount of land suitable for farming will decrease.
This means that efficient use of the land is critical – which in turn means an increase in the use of fertilizers. However, these bring their own environmental risks, as well as the direct hazards that over-exposure can cause for human health.
This means that when they are used, they need to be used as efficiently and accurately as possible.
How Is Artificial Intelligence Used In Practice?
John Deere has developed machine learning technology designed to ensure that where herbicides and pesticides are used, they are used as sparingly as possible.
Not only does this vastly cut down on waste, reducing the energy usage and environmental impact of pesticide production, it means that the impact of pesticides in the areas where they are used can be minimized. This means less pollution of local rivers and waterways from runoff, while also ensuring food production continues at optimum levels.
What Technology, Tools And Data Were Used?
John Deere uses technology developed by Blue River Technology, which it acquired in 2017.5 It harnesses computer vision techniques to se
nse where crops are threatened by pests, and control robotic equipment capable of firing accurate blasts of pesticide chemicals at the afflicted crops, while leaving others untouched.
Before being acquired by John Deere, Blue River Technology had built up a vast database of crop photographs. It then used computer vision algorithms to determine which photographs showed crops that were affected by pests and those that were clean or healthy. After being trained on this dataset, farming machinery was equipped with sensors capable of making the same decisions in real time while deployed in the field.
This machinery basically takes its own photographs of crops (in this case, lettuce), compares it with pictures of both healthy and afflicted crops, and makes a decision about which category to put each individual plant into.
Traditionally in large-scale agriculture, decisions as to whether or not crops should be sprayed are taken on a field-by-field basis, resulting in hugely inefficient usage of chemicals, which may have only been needed in a small area. This targeted approach has become known as “precision agriculture” and is only possible thanks to machine learning and computer vision.
This initiative is just the latest in a number of measures taken by John Deere to position itself at the cutting edge of artificial intelligence (AI). It also provides a service known as Farmsight,6 which allows farmers to make data-driven decisions about where and when crops should be planted. The data is crowd sourced from farmers all over the world and made available via subscription. The system gathers insights based on temperature, soil moisture levels, weather data, sunlight and many other factors to help farmers make decisions such as when and where to plant their crops to gain the highest yields.
What Are The Results?
Willy Pell, director of new technology at Blue River, said that their precision agriculture system has the potential to reduce the amount of pesticides sprayed onto land by farms worldwide by up to 90%.7 This should lead to less pollution as well as a decreased impact on human and animal health caused by the hazardous chemicals.
It will also mean higher crop yields for farmers, and help solve the challenge of feeding an ever-increasing number of humans with an ever-shrinking amount of available farmland.
Key Challenges, Learning Points And Takeaways
Advanced AI could provide solutions to the problem of producing enough food for the world's growing population.
Precision agriculture means a reduction in the amount of harmful chemicals sprayed on crops – increasing efficiency and reducing pollution.
Automation is not new to farming, but combining automated systems with advanced sensing and decision-making technology is helping to break new ground.
Challenges include teaching automated systems to recognize the difference between afflicted and healthy crops – this was done by training the systems on vast amounts of photographic data.
Notes
1Lightreading, John Deere Bets the Farm on AI, IoT: https://www.light reading.com/enterprise-cloud/machine-learning-and-ai/john-deere-bets -the-farm-on-ai-iot/a/d-id/741284
2NASA, How NASA and John Deere Helped Tractors Drive Themselves: https://www.nasa.gov/feature/directorates/spacetech/spinoff/john_deere
3Sentera, https://sentera.com/johndeere/
4United Nations, World Population Prospects: Key Findings: https://esa. un.org/unpd/wpp/Publications/Files/WPP2017_KeyFindings.pdf
5John Deere, Deere to Advance Machine Learning Capabilities in Acquisition of Blue River Technology: https://www.deere.com/en/our-company/ news-and-announcements/news-releases/2017/corporate/2017sep06- blue-river-technology/
6John Deere, Farmsight: http://www.deere.com/en_US/docs/agriculture/ farmsight/jdfarmsight_faq.pdf
7Wired, Why John Deere just spent $305 million on a lettuce farming ro- bot: https://www.wired.com/story/why-john-deere-just-spent-dollar305 -million-on-a-lettuce-farming-robot/
44
KONE: Using Artificial Intelligence To Move Millions Of People Every Day
Finnish-headquartered elevator and escalator engineering and maintenance group KONE are responsible for 1.1 million elevators worldwide.
It considers its mission to be to improve the flow of urban life, and at Heathrow Airport in London alone it is responsible for moving 191,000 people every day using 1,035 escalators, elevators and autowalks.
In 2017, KONE announced an ambitious data-driven program with the target of measuring and analyzing data collected from thousands of pieces of machinery all round the world. The information will be processed with machine learning algorithms and made available to other operators and maintenance businesses.
What Problems Is Artificial Intelligence Helping To Solve?
With so many moving parts across a large number of complicated systems, breakdowns and faulty equipment can mean that thousands of people are affected by delays.
Having to wait until things go wrong before remedies can be put into place leads to further downtime and inefficiency, while replacement parts are sourced and moved to the locations where they are needed.
As well as this, coordinating different people-moving equipment in large buildings is a difficult task. When someone presses an elevator call button, the system has to decide which car is best placed to respond. In many situations this won't be the closest one – which may already be full, or heading in the wrong direction. As this is traditionally handled by non-intelligent machines, it can often mean passengers waiting for longer than necessary.
How Is Artificial Intelligence Used In Practice?
KONE began the process of teaching machines to operate themselves back in the late 1980s, when microprocessor control of elevator systems started to become the norm. Processors were designed to estimate the average number of passengers that would be waiting at each floor and adapt the way they operated predictively.
Today, KONE has connected more than 1 million of its escalators and elevators to the cloud. They are fitted with sensors that can pick up everything from the start and stop times of elevators leaving and arriving at floors, to acceleration, temperature, noise levels and the frequency of vibrations running through cables.
KONE CEO Henrick Ehrnrooth told me: “We are connecting elevators and escalators to the cloud … that means we're connecting a lot of data, and this enables us to provide significant value for our customers.
“When you're managing a building, it's important to have a full understanding of what's going on, all the time – What is happening? How is the equipment performing? How are people moving in the building?”
With all of this data, machine learning algorithms are able to build models that enable correlations and outliers to be determined, which leads to a build-up of the machine's “understanding” of when faults or breakdowns are likely to occur. This means maintenance work can be more efficiently scheduled, and replacement parts are more likely to be in the right place at the right time.
Artificial intelligence also informs the “group control” function of elevator systems, which coordinates the way that multiple elevators operate together – for example, deciding which elevator is best placed to respond to a waiting passenger's press of the call button.
This is done by taking into account the predicted demand and availability of every elevator in the system together, and making decisions on the best way to move everyone efficiently.
KONE packages this data up into a service it calls KONE 24/7 Connected Services and sells it to other operators, enabling them to take advantage of machine learning-driven predictive analytics themselves.
What Technology, Tools And Data Are Used?
KONE launched its 24/7 Connected Services through a partnership that saw it working together with IBM. More specifically, it uses IBM's Watson cognitive computing platform to understand and learn how its machinery is working.
You can actually listen in on the system's conversation at http://machineconversations.kone.com – and experience for yourself what has been described as “both really dull and truly fascinating” int
eraction between machines.
Data is collected by sensors connected throughout the machinery, and this includes a limited amount of “edge computing”, where decisions about which data is or isn't useful are made within the sensors themselves. This helps to reduce the overall data volume by cutting out worthless “noise” at its source.
What Were The Results?
KONE, as well as other engineering and maintenance businesses using the connected system, is able to better understand the operation of its machinery and more accurately predict breakdowns and failures.
This cuts down on wasted time and energy – both on the part of the engineers themselves and the millions of people that rely on its equipment to get them from A to B each day.
On top of this, machinery can operate more efficiently. For example, an elevator can learn how busy it is likely to be at certain times of the day, and adjust the time it waits at each floor to allow passengers to enter. In buildings with multiple elevator systems, their operation can be co-ordinated so they travel more frequently to floors where they are needed, reducing passenger wait time.
Key Challenges, Learning Points And Takeaways
As buildings – and populations – grow larger, improving the efficiency of systems responsible for moving people around is vital to ensuring the smooth flow of urban life.