Amtrak’s Northeast Corridor is the busiest railway in North America, and also the most grumbled about. Its reliability is about average for the government-owned network — about one in five of its trains ran significantly late last year — but because it connects the centres of US financial and political power, it is the subject of the loudest complaints about slow service. Out of sight of the disgruntled passengers, however, Amtrak’s engineers are being supported by some of the world’s most advanced technologies to prevent those delays. Siemens, the German group that built some of the locomotives used on the Washington to New York line, is deploying what is known as the “industrial internet of things” to predict problems before they happen. By analysing data from 900 sensors on each of its locomotives, Siemens can understand why equipment failures occur, and recommend interventions that will prevent them in the future. Delays were down 33 per cent in 2016 from 2015, and performance on some measures is “almost an order of magnitude better than in the past”, says Rick Shults, Amtrak project manager for the Siemens locomotives. “They are able to introduce a concept for change well before we have even noticed there is a problem.” 1: Data on air pressure in the equipment, any impact on the strips and other information are collected and stored on the train 2: Each time the train is in a station, diagnostic information is uploaded through a wifi link. The data are collected at Siemens data centres in the US and Germany 3: Engineers at Siemens’ Data Analytics and Applications Center in Atlanta look for signs that carbon strips are about to fail 4: Messages are sent to Siemens and Amtrak staff at maintenance depots, telling them which strips need to be replaced and identifying what caused the damage Siemens’ work for Amtrak is at the vanguard of a revolution that promises radical change for industries including manufacturing and energy as well as transport. The plunging costs of sensors, communications, data storage and analytics have made it possible to record and process huge volumes of information about physical systems, from trains to oil refineries to wind turbines. Analysis of temperature, pressure, vibration, movement and flows of electrical current can be used to prevent failures, streamline maintenance, improve performance — and even change the way products are designed and made. By 2020, companies will be spending about €250bn a year on the internet of things, with half of all that spending coming from the manufacturing, transport and utility industries, according to the Boston Consulting Group. “It’s a huge opportunity for all industrial companies,” says Bill Ruh, chief digital officer for General Electric, the US conglomerate. “Data analytics and machine connectivity are the way to get to the next level of productivity.” Along with advanced robotics and 3D printing, the internet of things is one of the technologies that are expected to transform manufacturing over the next couple of decades. “I am not saying it can change: it will change, one way or the other,” says Roland Busch, chief technology officer of Siemens. “And there will be winners and losers.” Any industrial company that wants to still be around in 20 years is building up its digital skills and technology, often through acquisitions. GE last year bought four companies to strengthen its digital business. Honeywell this month bought an Israeli company called Nextnine, to reinforce its business offering internet security, a critical issue for industrial operations. Siemens has spent $15bn on US software companies since 2007, and has 21,000 software engineers. Germany’s Bosch says it has more than 20,000 software engineers, of which 4,000 are focused solely on the internet of things. GE has 14,000 software engineers, and is planning to hire 6,000 more technical and support staff for its digital operations. Branding consultants have found a rich seam of business naming the software platforms. GE has Predix, Siemens has MindSphere, France’s Schneider Electric has EcoStruxure, Zurich-based ABB has ABB Ability, and so on. Jeff Immelt, who this month announced he was stepping down as GE’s chief executive, has staked his legacy on making the group a “digital industrial” business, combining physical products with information technology. When he started his farewell tour of GE with his successor John Flannery last week, the first place they visited was the headquarters of the group’s digital business in San Ramon, California. The potential market is growing very fast. Last year there were 2.4bn connected devices being used by businesses, and this year there will be 3.1bn, according to Gartner, the research group. By 2020, it expects that number to have more than doubled to 7.6bn. Like Amtrak’s trains, however, the shiny digital future is arriving later than some had hoped. The potential is real, says McKinsey senior partner Venkat Atluri, but industrial companies have been slow to exploit it for a variety of reasons. They may need to change their organisations radically to benefit from the new technologies. Another obstacle is that there are so many different products and services available that industry standards have not yet emerged. Working with expensive and potentially hazardous machinery, industrial businesses are cautious about entrusting critical decisions to outsiders. “Customers are risk-averse, because they have to be,” says Guido Jouret, chief digital officer of ABB. “If you do something wrong, you can hurt people.” Potential customers are also very cautious about control of the data that reveal the inner workings of their operations. Gehring, a German company that makes machines for honing metal surfaces to very fine tolerances, is one of the showcase users of Siemens’ MindSphere digital platform. Wolfram Lohse, Gehring’s chief technology officer, says the carmakers that are its customers have been very careful about how it uses production data. Gehring has benefited from the new technology, Mr Lohse says, and is hoping for further gains including increased productivity from its machines. But he adds: “I have to admit it’s still just potential. It’s not in commercial use yet.” In this challenging and crowded market, new competitors are emerging all the time. Moving further into software brings manufacturers up against specialised information technology companies, and there is a shifting landscape of competition and co-operation between different groups that a decade ago could have safely ignored each other. Established companies including IBM, SAP, Microsoft, Intel and Cisco also offer predictive maintenance technology, and there are numerous start-ups seeking to exploit the new opportunities in industrial markets. As of last year there were more than 360 companies offering internet of things platforms, according to IOT Analytics, a Hamburg-based research group. Making money in that environment is not easy. GE says its digital business will not start making a noticeable contribution to earnings until 2019-20. For an example of how tough the competition is, take the contract for its “digital transformation” awarded last year by Engie of France, one of Europe’s largest power generators and gas suppliers. Engie talked to many companies about the contract to provide a single IoT platform for all its businesses worldwide, to use their data to improve efficiency and customer service. GE, which already had Engie as one of its biggest customers, might have seemed like the obvious choice to provide that platform. Instead, Engie announced in June last year that it had chosen a Silicon Valley company called C3 IOT, founded in 2009 as C3 Energy by Tom Siebel, a billionaire pioneer of customer relations management software. A month later, Engie also signed a partnership agreement with GE to work on a range of activities including improving the performance of its power plants, but that looked like a consolation prize. Mr Siebel describes GE as “a 19th century conglomerate [with a] 19th or 20th century vision of how to build software”, investing billions and hiring thousands of people to develop its products. “It doesn’t work like that,” he says. “You build great [software] products with 10 or 20 or 30 people.” C3 IOT had 130 employees in May. In November, GE agreed a deal with Exelon, the US utility, to improve performance at its plants and develop new applications for the electricity industry. It has other customers for its digital platform to manage power plant reliability and performance, including Sonelgaz of Algeria and Invenergy of the US, but the Engie deal is a sign of how dominance in hardware will not necessarily translate into software. The industrial internet of things in numbers Siemens' gas turbine factory in Berlin © Bloomberg €250bn Estimated amount that companies will be spending on the internet of things by 2020 7.6bn Estimated number of connected devices operating in industry by 2020, up from 3.1bn this year $15bn Amount Siemens has spent on buying US software companies since 2007. It has 21,000 software engineers The manufacturers argue that they have specialised knowledge of their industries and customers that will give them a crucial competitive edge. “We know how to manufacture. We know how to automate a building. We know how to run trains, or a turbine,” Mr Busch says. “So that means we cannot only provide the scale for software, but we have the domain expertise. This is the powerful lever we have, which no other company has.” The potential risks if something goes wrong also lend weight to sticking with well-established suppliers, says Mr Jouret of ABB. “Customers think: ‘Wait a minute: can I let you run my factory?’” But there are stories popping up that suggest the big manufacturers would be foolish to rely on the advantage of incumbency. Precognize, an Israeli predictive maintenance start-up, has won a contract from Germany’s BASF, the world’s largest listed chemicals company, to work at its plants. Chen Linchevski, Precognize’s chief executive, is happy to admit he and his team are no experts in chemicals. “We say: ‘We don’t know anything about your business, but the important part is that you do. You know about your plant, we know about software’,” he says. Offering specific services and letting customers keep greater control of their data is often more appealing than turning up with a “black box” and saying “don’t worry about it: trust us”, Mr Linchevski adds. The decision raises a fundamental question for all businesses: what is it exactly that they do? If they hand over control of their data, and defer to a service provider on decisions about how to run their operations, what value are they adding? Francesco Starace, chief executive of Enel, the Italian electricity group, says the company will buy from software suppliers, but will integrate their technology and develop its own services. There is agreement among executives and analysts that the present proliferation of competing platforms cannot last for ever. Zia Yusuf, a partner at BCG, says that in each segment of the market he would expect perhaps three or four dominant suppliers. It is possible that companies will be able to control segments where they have been traditionally strong: GE and Siemens in power generation, Siemens in manufacturing, Honeywell in refining, and so on. Mr Ruh says GE will be “one of only two or three companies” that can do the optimisation of equipment and operations that he sees as the group’s focus. But in such a new and complex market, the picture is constantly changing. “Five years from now, the approach will have evolved again and again,” Mr Starace says. Joe Kaeser, chief executive of Siemens, says it makes sense to be apprehensive about the competition the company faces. “Paranoia is not a good thing. Being scared is good,” he says. “We are sometimes a bit scared, and it utterly focuses our senses.” Gas turbines: How ‘scary’ AI runs system better than humans Some of the most radical changes resulting from the new world of connected devices come from using artificial intelligence to make decisions about complex systems. Gas turbines are a relatively clean technology for generating power compared with coal, but they still emit nitrogen oxides — pollutants that can cause smog and acid rain. Siemens has been using machine learning to find ways to reduce emissions, analysing data from thousands of sensors recording temperature, pressure, gas flows and other factors on every turbine. The results were that emissions could be cut 15-20 per cent, Siemens says, with the computer finding new ways to run the turbines that its human engineers had never discovered. “Our engineers do it from their experience, their domain know-how. AI does it in a different way,” says Roland Busch, Siemens’ chief technology officer. “Sometimes the system itself comes to a solution which you had never thought about. It’s a little bit scary.” Francesco Starace, chief executive of Italian electricity group Enel, says that experience can be replicated across the industry. The proliferation of cheap sensors and communications “will enable you to bring performance to levels that you would never think of, and maintain the plant in ways that you would never think of”, he says. The ability of AI to surpass human decision-making raises the question of how valuable a company’s specialised knowledge of its industry really is. Mark Hung, an analyst at the research group Gartner, says it would be wrong to conclude that human knowledge and ability are dispensable. “You are going to get new insights that you never thought of before. But it still takes a human to process that. Humans alone or machines alone are insufficient. Neither one is as good as both together.” As technologies for collecting and analysing data keep advancing, the importance of specialised human and corporate knowledge about an industry may increasingly be called into question.