Two years ago, tech giant Intel launched a proof-of-concept (POC) technology test at its manufacturing facility in Penang, Malaysia. The goal was to ascertain whether collecting and analyzing data from sensors linked to manufacturing equipment could deliver much-ballyhooed promises around enabling predictive maintenance to boost production yield and reduce operational costs. That POC evolved into a full-blown pilot, which Intel, last month, said generated $9 million in cost savings.
Now, Intel is looking outside the four walls of that factory, and is collaborating with Mitsubishi Electric and other partners to offer its IoT expertise, in order to help other businesses improve factory automation through a combination of hardware, software and consultation. In many cases, the legacy equipment within manufacturing facilities is already collecting basic diagnostic data or images that can be leveraged for performing predictive analysis or improving efficiency. IOT Journal spoke with Shahram Mehraban, who manages Intel’s industrial automation and smart grid segments, about the pilot and the data analytics capabilities that translated data into dollars saved.
The pilot began a couple years ago, Mehraban says, and was preceded by a proof of concept. “I come from the business group, and we are responsible for [managing] profit and loss,” he explains. “We looked at the technology and manufacturing group as a customer. So the goal of the POC was to show the stuff we were talking about with regard to IoT in the business group was real.” That is, he wanted to prove that data collection and analysis could help improve factory automation.
For the POC, data analytics helped to predict when a pick-and-place machine used to build central processing units (CPUs) required repair or replacement. Over time, Mehraban explains, these machines begin to misalign, which can damage the boards, resulting in their needing to be scrapped. “Depending on the type of CPU, this can be very expensive if we have to scrap it because of this damage,” he says. The factory can simply replace the pick-and-place device periodically to avoid such errors, he notes, but adds, “these devices are very expensive, so we want to avoid the unnecessary replacement of parts.”
The POC showed that by tracking the number of relays made by the pick-and-place machine, Intel could predict when the device would begin to misalign—and then repair or replace it before it can begin damaging products.
Pleased with the preliminary results from the POC, the manufacturing group greenlighted a full pilot consisting of three use cases, each of which leveraged the C Controller from Mitsubishi Electric’s iQ-Platform. This gateway, which employs Intel’s Atom processor, collected data from existing meters and cameras used at the manufacturing facility. The gateway also plays an important role by translating data, presented in various formats based on the sensor that collected that data, into a common protocol. In some scenarios in which a use case requires real-time decision-making, the gateway performs some data analytics as well. From the gateway, the data was hosted on Dell‘s PowerEdge VRTX on-premise server, and was processed using Revolution R Enterprise software provided by Revolution Analytics, hosted on the Cloudera Enterprise database service. Intel and its partners say they can develop IoT applications by leveraging existing infrastructure inside factories, as well as by collecting, processing and analyzing data—enabled through the gateways, servers and software they provide.
Intel’s pilot program consisted of three use cases. One focused on monitoring automated test equipment used to spot defects in products being manufactured. Over time, this unit begins to wrongly identify good products as defective. During the pilot, the gateway communicated the test data to the factory’s data center onsite, where the collected information was run through an analytics model developed by an Intel partner to determine each component’s “time to failure.” By counting the number of times the automated test equipment is used and predicting when it might begin to fail, the software successfully foresaw up to 90 percent of the failures before the factory’s existing online process-control system could detect problems with the equipment.
The second test case involved equipment used to place tiny balls into solder paste on a substrate that then move them through a reflow oven. The balls, used to connect electronic components, are placed via a vacuum, but occasionally, due to changes in vacuum pressure, the balls can become misplaced or end up missing from the solder paste. For the pilot, additional sensors were placed on the vacuum unit in order to measure pressure levels. This data, along with images from the existing machine-vision system (which takes a photograph of each substrate following ball placement, to determine whether each ball was accurately placed), was analyzed to ascertain how changes in vacuum pressure correlate with faulty ball placement. Again, this resulted in preemptive alerts being issued to the maintenance crew, who could then quickly adjust the pressure to avoid significant losses in the factory yield.
In the third use case, analytics software replaced manual inspection steps using high-resolution pictures of products to determine quality. Prior to the pilot, the imaging software first separated the products into “good,” “bad” and “marginal” categories. Engineers would then manually inspect the marginal goods under a microscope to decide whether they should be scrapped. “We needed a highly automated tool to look at marginal products,” Mehraban explains. This came in the form of analytics software that, according to an Intel case study, converts each image into data that the company can employ to differentiate true rejects from marginal ones.
According to Mehraban, the case studies quantified the value of IoT applications within the manufacturing setting, and are helping to convince other companies of the technology’s value in tangible ways. “There are a lot of white papers talking about how manufacturing is the biggest beneficiary of IoT,” he says. “Everyone gets the basic premise of turning your information into actionable data, but when it comes to implementing IoT, it feels overwhelming.”
Mehraban says Intel and its partners—Mitsubishi Electric, Revolution Analytics and Cloudera—are presently in talks with a number of manufacturing companies. “We’re looking at how we can make the transition to [leveraging the IoT] easier, and are providing blueprints [to do so]. It’s not going to happen overnight—even Intel does it step by step,” he states. “But we’re in talks with some manufacturers who have been hearing about how Intel is eating their own dog food when it comes to IoT.”