How integrating more sensors into robots will help increase up-time, optimize maintenance schedules and enable new business models.
Nothing is quite as frustrating as a machine that unexpectedly stops working, such as our car failing to start off one morning. In order to avoid this nuisance, we should stick to the recommended servicing intervals. Nevertheless, this practice is not really desired, because one often has to spend additional time and money on spare parts and services.
So, how can we find that balance between too frequent and too little maintenance? And how can sensors help us to optimize the scheduling of such maintenance?
In industry it is a similar situation: The automation equipment and robots employed can only get work done when they fully function. In order to do so, they need to be maintained regularly. However, turning these machines off for performing maintenance takes away valuable time from production activities. In addition, automation teams might be replacing parts which are still fully functional because of the specific load profile within the production process. This causes significant cost which could have been avoidable, in case users of the machine had known better.
With automated machines and robots now commonplace in most factories, it would be easy to think that they would run until the electricity is turned off. However, it falls upon technicians and engineers keeping everything up and running.
Like our cars, most manufacturing equipment comes with some sort of servicing schedule. According to statistical service predictions, after a set number of hours of use, the machine will need to be oiled, have parts replaced, and important elements will require inspection. Sometimes, these inspections will be fast and simple. Other times, they may require many hours of work and (sometimes) expensive parts are to be replaced in order to avoid any later unexpected break-downs.
Depending on the type of usage, the intervals between these scheduled services could vary. A robot arm that regularly lifts one ton cast metal parts is likely to suffer more wear than the same robot arm that only lifts half that mass or moves much slower.
The maintenance team has to try balancing preventative maintenance activities, so that the equipment is available as much as possible at lowest possible servicing cost. At the same time, it should occur often enough that the machines don’t suffer a break-down during use. This would result in unexpected maintenance and lost manufacturing time.
Most organizations implement continuous improvement plans (CIP). Teams review failure and maintenance reports and discuss ways to avoid that issue occurring again. Once the team finds a solution, it is documented and implemented. Should the issue occur again, the documented resolution can be reused or modified.
These types of programs have become so effective, the failures found are beginning to cost more to rectify than the costs resulting from the issue.
The collection of the data required to recognize such issues is also becoming easier. As a result of Industry 4.0, equipment in factories is being gradually networked. This produces more data for analysis. If the scrap parts generated by a machine start to rise or the time taken to complete a process step slowly increases, this can be quickly seen. Such changes could indicate that the machine is in need of maintenance.
Often, we meet people who seem to be conjoined with their machines. Cars, power tools and office equipment – they will notice a change in the sounds it makes and state that the equipment will shortly fail. And then, within a few days, it does break down, indeed.
Such human intuition relies on our senses, experiences from the past and an understanding of other depending factors, explaining the possible root causes of the observed changes within the system.
Similarly, robots provide us with a wide range of signals that we can use to infer their health. Perhaps, a change in their whirring sound or a slightly juddering motion – these are all signs that a motor, gearbox or bearing is past its best. For instance, if power consumption goes up slightly but steadily, perhaps there is some increased mechanical resistance. This, again, can result from close-to-fail bearings causing the electrical system to draw more current than usual.
Sensors are used in large quantities in manufacturing today. Without them it would be impossible to ensure that temperatures and pressures were correct. It would be impossible to detect that objects are located in the right position at the right time. Optical and magnetic sensors are used to detect objects running along conveyor belts. Increasingly radar and laser sensing is used by autonomous guided vehicles (AGV), allowing them to navigate around the factory floor and to avoid any unexpected obstacles.
But their core use remains controlling and monitoring the manufacturing process. If a factory uses Industry 4.0 data collection, sometimes the data collected is only used to analyze a failure after it has happened.
Highly accurate sensors have become increasingly common, their prices falling to just a few cents each in some cases. The semiconductor industry has used the unique properties of silicon to create tiny sensors that measure temperature to accuracies of a tenth of a degree. Capabilities, such as micro-machining, have allowed mass-manufacture of chips featuring moving parts that are finer than the human hair. Such devices, known as MEMS sensors, are used in the creation of microphones.
Well-known physical effects, such as the Hall effect, are used to turn magnetic fields into electrical signals. Such sensors are used to measure how much current is flowing through a conductor. Other methods, such as giant magnetoresistance, use quantum mechanical effects that can be used to measure magnetic fields. Implemented as a Giant Magneto Resistance (GMR), such sensors can be used to measure the speed of motor rotation or the angle of a mechanical shaft.
In some cases, several sensors are all integrated into the same device. This is often the case with temperature sensors. Temperature measurement devices are very simple and can be easily integrated alongside the analog electronics of other sensors.
With this wealth of low-cost but high quality sensing capability, there is little reason not to integrate a range of sensors into robots to monitor their health. For example, temperature, vibration, noise, positioning and motion, acceleration and force sensors could be integrated into the joints of a robot. Many of them can be found there already, as they are required to enable the intended function of the robot or any other machine. But in addition monitoring all these sensors together would, like an intuitive human, recognize signs of potential early failure. An increase in average temperature, combined with juddering motion and higher power consumption, could highlight pending bearing or gearbox failure.
There are two key challenges which are associated to sensor based health monitoring and predictive maintenance measures:
Today’s manufacturing sensor may detect a few tens of objects every minute, in itself a significant amount of data. However, a set of robot motion related sensors integrated into a joint would be collecting thousands of data points every second. If each robot has four or more joints, this quickly becomes insurmountable.
The second challenge is the always existing additional risk that more sensors, sometimes not necessarily required for the operational algorithms of the machine, could fail and cause false alarms.
If this continual sensing is to be of use, it also needs to be evaluated immediately in order to highlight pending failures. Microcontrollers, such as the XMC4000 series from Infineon, can be used to collect, pre-process and evaluate the data. Featuring all the necessary data interfaces to connect to such sensors, these tiny number crunchers can also be integrated inside robotic joints. This fusion of sensor information can be evaluated on-chip or passed on to a central computer using Industry 4.0 data networks. EtherCAT, one such networking standard, is natively supported by the XMC4000 family.
Regarding the risk of false fails, it is important to use only sensors with highest quality standards combined with ad-hoc evaluation of the measured parametric data within a given context. Most of Infineon sensors fulfill even most demanding automotive quality standards and have successfully proven their longterm stability far more than a million times in cars running on our roads.
AI is all about pattern recognition and can be used to replicate something that looks like human intuition. Initially, the AI would be taught the expected fused-sensor output under normal conditions on a well-maintained robot.
By continuously reviewing temperature, loading, power consumption, and motion, the AI could later easily highlight when some of these parameters are starting to fall outside their expected range. This would be the trigger for the maintenance team to inspect the robot for excessive wear.
Manufacturing facilities require enormous financial investment both in their creation and maintenance. And maintenance teams work hard to keep such facilities in top working order. In many cases, it is often difficult to justify the cost of upgrading existing sites to Industry 4.0 networked solutions. However, with the availability of miniature sensing technology and powerful, but tiny microcontroller solutions, future robotics and manufacturing equipment is likely to deliver worthwhile insights.
As the power and integration of AI devices moves forward, robots and manufacturing equipment will be able to detect their level of wear themselves. If this can deliver improved uptime, and leads to more efficient factories, it will be the tiny silicon sensing and computing solutions that we will thank for improved business margins and profitability.
With the introduction of condition monitoring and predictive maintenance solutions into robotic system new advanced business models would become possible. Especially collaborative robots, which can be easily programmed and adapted on a particular task become increasingly interesting whenever a peak-load situation occurs in certain handling tasks within simple production or assembly processes. So far, a production team response is to expand workforce with more temporarily employed workers.
In the future a temporary lease of a fleet of (collaborative) robots might do the job. Wouldn’t it be nice, if the real usage profile of the machine could be assessed afterwards and the final bill could reflect this in pricing? Or if there was good inside knowledge whether the machine has been operated in constant overload during the rent causing accelerated aging? Use of sensors and powerful data processing could provide this extra functionality, which is about to become reality in the era of Industry 4.0 and (collaborative) robotics.
Last update: July 2018
