Unplanned machine failures. Unexpected crashes on the stock exchanges. A sudden heart attack. People everywhere are looking for ways to detect these risks early, to prevent them, or at least to optimally prepare for them. “Predictive” (i.e., anticipatory) is the buzzword companies in all industries and private individuals rely on. The basis: data.
Let’s start with Nostradamus, a predictive superstar among the prophets. French pharmacist, astrologer, and visionary, he predicted little good for 2018: unusual weather phenomena, floods, drought, and heavy hurricanes with devastating effects. Not to mention, a big war ahead that will divide the superpowers.
Whether you believe the prophecies of Nostradamus or not, one thing is certain: He has often been wrong and many of his predictions are so general that they inevitably happen sooner or later. Interestingly, many people are fascinated with the predictions of prophets, fortune tellers, and astrologers. They seek to prepare for and arm themselves against what may come in the near or distant future. But very few believe specific predictions because they lack the numbers, data, and facts.
Today, however, there are plenty of numbers, data, and facts. And they are constantly increasing in volume. Credibility is increasing. The forward-looking knowledge that will befall us, our companies, and our machines is based upon facts. Thanks to sensors, the Internet of Things, and new methods for comparing recent data with information from the past, future changes – whether short- or long-term – can be anticipated with increasing accuracy.
A prophet in the machine room
Example weather forecast: Today, the accuracy of the forecast for the next six days is as good as the forecast for the next day 50 years ago. The German Weather Service uses satellite information, data from hundreds of floating measuring buoys and national meteorological services, thousands of merchant ships, commercial aircraft, and weather stations, for example. High-performance computers process this flood of data and update the weather forecast every few hours. Outdoor pool or sauna: Accurate weather forecasts determine how we plan our leisure time. In this case, they fall under the lifestyle category. For farmers and shipping or logistics companies, however, the weather can be existential. As such, the use of weather services has been standard ever since they have existed.
“Self-learning algorithms and analyses identify patterns and dependencies of various operating parameters.” GEORG RÄTKER, Global Delivery Unit Automotive & Manufacturing Solutions, T-Systems
In industry, however, short-term forecasts are new, because the technological conditions first had to be created. And today they exist. The machines themselves acquire status data. In addition, sensors read data and send it via data networks to the cloud, where special algorithms process it together with historical data, aggregate it, and make it understandable for every machinist. And all in real time if desired.
“Self-learning algorithms and analyses identify patterns and dependencies of various operating parameters. On this basis, forecasts of failures and their causes can be made. This allows companies to determine machine and robot downtimes days in advance, to plan maintenance while optimizing resources, and to adjust the production process,” explains Georg Rätker, T-Systems expert for automotive and manufacturing solutions.
REDUCED UNPLANNED DOWNTIME
Prevention is better than repairing: Everyone will subscribe to this truism. But what are the actual benefits for companies that rely on predictive maintenance for their machinery? They want to minimize unplanned downtime and thus reduce costs. After all, the loss of production caused by stationary machines alone costs the automobile industry around one billion euros a year. And the costs of operating and servicing capital goods like conveyor and production equipment make up 80 percent of the total cost of ownership of capital goods such as the former. The acquisition costs, however, account for only one euro of every five.
Depending on the industry, maintenance costs are up to six percent of the total cost of an industrial enterprise. In addition: “Businesses in Europe are not overly optimistic about the efficiency of their industrial equipment and vehicle maintenance processes. A significant majority of companies consider their maintenance processes to be not very efficient,” confirms Miloš Milojević, industrial analyst at PAC.
Weighty arguments. But according to a November 2017 Frenus poll on the status quo of the mechanical engineering industry in Western Europe, more than three-quarters of companies still do not use predictive maintenance solutions. “At the same time, however, a clear majority believes that the use of such solutions in the long run is vital for the survival of mechanical engineering companies, since they will help to survive against competition from low-wage countries,” says Rätker.
CURRENT MAINTENANCE CONCEPTS OUTPACED
The consulting firm BearingPoint also found out that predictive maintenance is rather discussed than implemented. While 84 percent of respondents have already addressed predictive maintenance in their organization, only one out of four have completed their first projects,“ although traditional maintenance concepts no longer meet today’s needs,” says Donald Wachs, Global Director of Manufacturing at BearingPoint and an Industry 4.0/IoT expert. “They tie up capital and consume too many resources.” With predictive maintenance, however, four out of five companies want to increase equipment availability and reduce maintenance costs by 60 percent.
So why this restraint when the benefits are obvious? The supposedly high costs of expensive solutions and sensors are one reason. BearingPoint also discovered that the majority of companies (57 percent) see IT security as the biggest technical hurdle. And 61 percent fear the high costs of implementation. There is also a lack of courage to make mistakes and learn from them, says Donald Wachs. “This inhibits companies and blocks the potential of predictive maintenance.”
In today’s manufacturing industry, IoT modules are already being factored in production planning.
Lack of data, according to Britta Hilt, Co-Founder and Managing Director of IS Predict, is another reason companies are having a hard time with predictive maintenance. The software provider has been developing solutions based on self-learning artificial intelligence and predictive data analytics for nearly a decade. “We face the typical problem of different innovation cycles,” says Hilt. “Many machines have been running for ten years or more, so they are not equipped for today’s analytical capabilities.” While most machines capture process and product data, these data oftentimes cannot be read out for predictive purposes. “New machines, on the other hand, have an increasing number of IoT modules that can capture and transmit status data. It will take some time for all machines and equipment to be equipped for predictive maintenance from the outset,” predicts Hilt, who has been involved with predictive intelligence for more than 20 years.
The potential, however, is great. For example, IS Predict has used predictive maintenance to reduce unexpected failures of locomotives of an international logistics company. For the analysis of the machine data, IS Predict taps up to 900 data points for each of the more than 4,000 locomotives. Using eight parameters, specialists were able to develop a characteristic pattern that indicates when engine failures occur, thereby reducing the number of engine failures. Today, the logistics company detects failures up to three weeks in advance and proactively maintains the affected locomotives without failures. This saves money, since the cost per engine change incurred until now was around 200,000 euros.
To minimize the maintenance and downtime of locomotives, up to 900 data points are tapped on each railcar.
Next Generation Maintenance
Another example: With IS Predict, T-Systems has developed a process in automobile production that can optimize the maintenance of welding robots. If a robot fails unexpectedly, the entire production line shuts down. And, according to a survey by market research firm Nielsen, this costs the company up to 22,000 US dollars per minute. Today, the T-Systems customer identifies failing welding robots up to six days in advance and also recognizes the causes of the impending defect. Spare parts that fit those of the OEM are available and maintenance is scheduled during production breaks.
The automotive industry alone loses one billion euros a year due to production downtime caused by idle machinery.
T-Systems goes a step further than predictive maintenance with their “Next Generation Maintenance” approach. In this case, however, data is not just processed for the purpose of reducing downtime or optimizing service: “We link the machine data from maintenance back to engineering,” says Rätker, describing the new approach. “Weaknesses in production equipment identified during maintenance are incorporated into the further development of a machine. An analysis of the maintenance history provides the necessary conclusions. “In this way, the quality of the products can be optimized, and the service life of a machine or entire production plant extended. Rätker: “This can significantly reduce the total cost of ownership.” For example, if a machine breaks down more frequently at certain temperatures, the producer can change the material composition at the critical weak points. Or the company ensures the temperature in a production hall is optimally adjusted.
Every year around four million cell phones disappear in Germany. With every theft or loss, the owners face expensive replacement costs. A blockchain for smartphone identification numbers could facilitate the locking of mobile devices.
Interview with Donald Wachs, Global Director Industrial Manufacturing and Industry 4.0 at BearingPoint, about predictive maintenance in the industrial sector and why smaller use case projects take longer to get you anywhere.