 Don't you wish you knew in advance when mechanical stuff was going to break? I mean, I really didn't intend on flooding the neighbor's condo downstairs, and my dishes didn't even get clean. But mechanical systems are always giving us clues about when they plan to fail. Their vibration changes, they make slightly different sounds, their temperatures rise. The hints are all there. Our gear is talking to us. It's just we that aren't paying attention. Hi, I'm Amelia Dalton, host of Chalk Talk. In industrial systems, it really pays to pay attention. If we monitor key factors that can give us early warnings of failures, we can do predictive maintenance which can dramatically improve uptime and eliminate costly failures. In this episode of Chalk Talk, my guest is Manuel Cantone from ST Microelectronics. And we're going to talk about Sensor Tile Wireless Industrial Node, a new integrated sensor solution that makes industrial monitoring a snap. And before we get started, don't forget to click that link. There you can find even more information about ST Microelectronics Sensor Tile Wireless Industrial Node. Hi, Manuel. Thank you so much for joining me. Hi, Amelia. Okay, so we hear a lot about industry 4.0 these days. But Manuel, what were the first three industries? Thanks for asking, Amelia. So this is definitely a hot topic. And we refer actually in ST to industry as smart industry. The 4.0 terms comes when you acknowledge the existence of the previous industrial revolutions. These are the major release of industry when something dramatically changed in the way that men are producing products. So I have the industry 1.0, so the very beginning of industry when mechanization and steam power introduced a completely different way to produce things rather than end producing everything. Industry 2.0 is where mass production came into place. And this is thanks to assembly line and electrical energy. Good example is Ford with its assembly line. Industry 3.0 is when the first computers and electronics are added in the mix. And basically can automate manufacturing and can control a specific machine to read with electronic equipment. So what are we adding to all of these with industry 4.0? We are adding the connectivity piece to it, the IoT, the Internet of Things. So we basically allow every machine and every component to now communicate to the cloud, to a bigger entity, and to build in this way cyber physical systems. And we will see later on what are the implications of that. So Manuel, what kind of opportunities are we really looking at at industry 4.0? Industry 4.0 will allow our customers to produce more efficiently and in a more environmentally friendly manner to respond to demand in a more flexible way and sometimes even with the customization and ultimately to provide a better and safer human experience. An immediate opportunity that is, if you want to allow and give fruit here is trying to optimize the maintenance of equipment, taking advantage of the data that are available now from the equipment itself. Okay, so let's talk a little bit more about predictive maintenance, Manuel. Absolutely, and it's a bigger, bigger opportunity. So the value proposition of predictive maintenance is allowing to shift from a preventive maintenance approach. Think of our oil change in the car. So we basically schedule a maintenance every three months and whatever the status of the car is, we do change things and we do spend money on it and do something. We try to move away from this for very expensive equipment. And the first step is to move to a condition-based maintenance. So there is an inspection prior to a maintenance itself. And this is of course condition per se. Right now I'm referring to the fact of reading on the spot data coming out of the machine. So an operator goes, take a look at the status of the machine and decides if a maintenance or if some operations are required or not. Think about your HVAC system. That's what we do in our home with our heating and cooling. The operator comes and if he needs to do something, he does. Otherwise, he just showed up and you pay for the truck hold. What we are trying to enable here, adding the connectivity to the sensing, is to have the machine continuously observe certain parameters and being able to trigger an alarm on a specific threshold, on a specific behavior that is happening, to then trigger the maintenance. So the idea is basically that the machine is called and called for maintenance whenever it's needed. And as you see in an optimized way. So in this way, I can move a step forward from condition-based maintenance, adding the connectivity piece. And here's where industry 4.0 really takes place. Minimizing the downtime and basically maximizing the production time of the machine itself, reducing the risk. The machine breaking down because I'm continuously observing what's happening rather than at any given moment in time. So Manuel, I can definitely see how predictive maintenance could mean a lot in this situation. But let's talk about retrofitting. Now, I know there are a lot of facilities out there that need help getting to 4.0. That's absolutely a good point. And all of these facilities, they're not willing to basically retool all their production with new machines. So retrofitting is definitely a good topic. And from ST's standpoint, we want to provide a reference design, a smart sensor reference design that will ease our customer journey into industry 4.0. And basically take this sensor reference design and see what they can do and what information they can extract from their own existing machines. So Manuel, what do I really need to do to implement predictive maintenance? That's a good question. I mentioned earlier cyber-physical systems. So what does that mean? So I start with my factory floor, where the existing equipment is. And I'm fitting either retrofitting or natively fitting in many machines with the sensors and then connect the sensor somehow to a centralized intelligence to the cloud, where I can do data analysis and I can take all the sensor data coming from my machine and identify patterns and identify basically symptoms for earlier failure. From these centralized intelligence, I also run my ERP systems. And I can basically schedule a call or send an operator and close the loop from the physical world to the machine. So starting from the sensor, going all the way to the cloud and then go back to the physical world with, for example, an operator call. So Manuel, you've been referring to the edge. Now, are you talking about edge computing here? Yes. As you see, I encompass the edge here between the smart sensor and the gateway. And definitely, we are talking about edge computing. And there is a concern about closing this loop between sensors and cloud, concerns about latency, and sometimes even concerns about privacy. So how to move that decision-making to the edge? And the gateway is the first entry point to the cloud, where I need to be able to do some processing. I need to be able to run functions and that the cloud is capable to maybe to create alerts and to create some actionable alarms on the data that are collected. So I see you mentioned AI on the smart sensor node. And tell me a little bit more about that. Yeah, that's a good question. We put AI here all over the place. And what we are trying to say is that we can distribute that intelligence and not have the machine learning inference run all on the cloud. I can run functions on the gateway, as I was mentioning earlier. But I can also run function on what I define here as a smart sensor node. This smart sensor node are usually built with some kind of sensors, some kind of connectivity and with a microcontroller unit that is able to process locally the data and to act on the data. In ST, we provide the tools to run machine learning model inference on the microcontroller itself. It's called the STM32QBI. And thanks to our technology in sensors, you can also move part of that inference all the way down to the sensor where there is a machine learning programmable core that can take immediate action. And in this way, I addressed the latency problem that I was mentioning earlier. And I can also address at the same time a power consumption problem. I cannot have the sensor being alive and sending data all the time, but waking up on a specific event and then only then act. That enables battery operated smart sensor node and can enable retrofit, as we were mentioning earlier, an existing machine with the minimum impact possible on the machine itself. So Manuel, explain to me how exactly do these various sensors fit into this predictive maintenance picture? So the idea is that when a machine condition starts to change, it will manifest itself with some kind of signature that if I apply the correct sensor and I will be able to capture. Usually, if I take, for example, a motor, an electric motor or a bearing, the first things that change is the ultrasound signature of the machine itself. So if I'm able to sense that with an ultrasound microphone, then I will be able to capture a condition change months from the actual failure. Vibration monitoring is a close second. And if I do vibration monitoring at higher frequencies, I will be able to still capture the failure at the very early stage. Now, if I move into the graph, if I can observe a power consumption change in the machine, I'm probably down weeks from the failure itself. Then the machine, like our home cars, will start to do some noise. And then you know that there is something going on and that you are weeks at this point from the failure itself. Then if there is a temperature change, big signature that something is about to go bad, and if I see smoke, my machine is basically done. I need to, as fast as possible, shut everything down and try to see what's happening. So the idea is that adding sensors, adding all these sensors technology, which gives me more eyes and ears on what is actually happening to the machine. Yeah, Manuel, smoke is never a good thing. So let's talk about that vibration analysis first, which was one of the first sensors you mentioned earlier. So what exactly does that scenario look like? So by vibration here, I mean sensing inertial movement of the device from the five kilohertz down. And we have specific sensors that are able to do that. As I put in this graph, the IIS3DWB is a sensor that is specifically designed for vibration analysis. This sensor is a little bit more power hungry than our usual accelerometer or 6S, and that's why we have also another couple of options. And the idea is that basically you can start the sensing at high frequency based on something that is happening at low frequencies. To complete the spectrum at higher frequency, I need a different means of sensing. And here I specify two acoustic sensors, two microphone. One digital microphone that has limited to audible range and an ultrasound microphone that is capable of sampling all the way up to 80 kilohertz. Depending on the machine and the use case, you will need to apply a mix of the two sensing technology, the acoustic and the inertial or maybe all of them. For example, for fan bearings, they have a high signature at higher frequency. So definitely ultrasound for motor imbalance, there's a signature at low frequencies as well. And one mention here that our sensors here, specific for industry, I have a 10 years longevity commitment so that people that build with our sensor know that they can buy them for the next 10 years so they can definitely count on ST for having these sensors available for their machines. Okay, so Manuel, if I'm working on an IoT design today, how exactly can you guys help me? Can you help me through the whole life cycle of my project? Absolutely, yes. We understand that just buying a MEMS sensor is not the response to our customers need, but that is our core business. Selling components that you can find on Mauser is definitely what we want to focus on. But to enable and to give customers an idea on how to use our parts, we try to lower the barrier to get started providing development kits. These development kits come either in two forms, either stackable boards with modular software, our nuclear and ex-nuclear approach where I buy a microcontroller board and I add a functionality, for example, a vibration sensing functionality through a MEMS sensor board. But we also saw that to better teach our customers how to use our sensor and better visualize what are they capable of, we need to provide to the market also form factor boards. These are really useful and they help spark the imagination of our customers in a multitude of use cases. And the form factor boards are only one computing board. There is a microcontroller, there is a sensing, there is some mean of connectivity. Sometimes they are better operated and together with those, we provide firmware to get started and to run on and so application example and a full development ecosystem that they can use. So from code generators to development environment, the artificial intelligence toolbox that I was mentioning earlier to be able to run their machine learning model inference on the microcontroller or on the EZCO of the sensor distributed from the two and all the tools that you need to debug and to simulate the application. We then provide integration with cloud providers. We'll see an example of that later where we don't play with partner. And for example, we have a partnership in place with AWS in this particular case that provides all the services and tools that you need to build an end to end. In our case, predict the maintenance application starting from the sensors and going all the way closed loop. So Manuel, you mentioned earlier a dev kit. Now is there something specifically for industry 4.0 there? Yes, we just released this sensor tile wireless industrial node that is specifically designed with in mind condition monitoring and the predict the maintenance application is a part of our form factor sensor node. So as you see, 50 millimeter by 50 millimeter and it has a full set of ST technologies. So from the sensing, we have industrial grade sensors for variation analysis, sound emission up to 80 kilos and environmental monitoring. We have various options of connectivity. So there is a BLE module on the main PCB but we also provide Wi-Fi connectivity and modular expansion to LTE, et cetera. We have local processing with our STM32 L4 plus. It's a ultra low power ARM Cortex-M4 microcontroller. We have also a secure element footprint to all the decredentials of the device and to basically help you to connect securely to your cloud of choice. And all of these is powered also by ST components. We have linear regulators, switching regulators, battery chargers. So it's a good reference design and it's a good sample of all ST technologies and how to use them to build an industrial grade device. I mentioned earlier the 10 years commitment. So these applies to the sensor and the microcontroller in this case. And that's what makes it really specific for industry. So Manuel, what are those two gears about right there? I'm intrigued. So the two gears refers to the capability of this sensor node, of being a smart sensor node, of being able to run algorithms but also run machine learning model inference on the microcontroller. We also have a machine learning core on our sensor, our 6-axis IMU in this development kit and that machine learning core enables distributing the inference from the microcontroller and the sensor itself to have the least possible impact on the battery life of the device itself if we are referring to a battery operated node as this one is. Okay, so Manuel, if I ordered this kit from Mauser, what all will I get? So when you order the kit from Mauser, this is a full development kit. It comes with a debugger, so our ST-Link V3 Mini. So, and you see a cable over there also that is used to connect to the board. So we basically, you have access to all the IDE, the idea of your choice to program the device to your liking. You will find in the kit also a case and a battery so that basically you can use these to start retrofitting your equipment and start understanding what sensors you want to use. This case comes also with two fly-in for ease of connection and what you will not find on the kit is the Wi-Fi expansion. It's a separate part number as you see later but it's also available on Mauser and that if you want to add Wi-Fi capability and you will see all the sensors that are on this PCB on the graph on the right. So the microphones, the inertial sensor, the environmental sensor and the deposition of the BLE module and the STM32L4 plus. Okay, so keeping predictive maintenance and conditioning monitoring in mind, what kind of sensing solutions do you guys have for me here? In this development kit, we tried to put all the sensor we mentioned in the graph that was ending up with the smoke in this development kit. So we mentioned ultrasound and we have an analog differential microphone on the kit. We have also a digital microphone to sense the sound. We mentioned vibration and we have the ultra-wideband accelerometer IIS3WB and also other inertial sensors. We have the six-axis IMU with machine learning core and an ultralow-power accelerometer also to basically save battery life but keep on monitoring what's happening to the system. We have also a low-noise low-power magnetometer here and a full set of environmental sensors. Here I'm showing pressure sensors and temperature sensors. Okay, so this is all part of an overall predictive maintenance platform, right? What all were we looking at there? So to build a predictive maintenance platform, a development kit is not enough. We mentioned a whole closed loop with cloud application. So we have a development kit here that has the capability of connecting directly to the cloud. If I take this STWIN and add the Wi-Fi expansion board, I can connect directly to your cloud of choice in this case, we have AWS. And I can send the messages directly to the service that is able to listen to the sensor messages. If I use the firmware that I show here, the function pack industrial predictive maintenance one, the connectivity is embedded in that firmware. So you have that capability. To close the loop, ST also provides a sandbox. So this is a dashboard. It's called Dash Predictive Maintenance that is built with AWS services and we provide to our customers free of charge. Ask customer can connect their own MyST.com account to this dashboard, deploy up to five sensors and play with the sensors up to six months and collect data and basically do this end to end data gathering from the sensors to the dashboard itself. Manuel, what all can I do with this dashboard? It's super cool. So with this dashboard, you can basically put together a real POC. You can connect up to five nodes and you can add the geographic location, for example, to these nodes. And you can add threshold on certain parameters so that alarms and warnings are triggered from certain sensor data. And you can see the live data being streamed to the dashboard section of this cloud application where you can see real time the sensor data coming from the environmental sensor, from the vibration sensor and for the ultrasound. Vibration and the ultrasound go through a fast Fourier transform that is performed on the micro controller so that the data flow is minimized and optimized for transmission over Wi-Fi. Okay, so I can see the data, but Manuel, what's next? That's an excellent question. And next is moving from condition-based monitoring to predictive maintenance. We store all this data into a data lake and allow our customer to download this data onto their own machines. And what we expect and what we propose is for our customer to analyze this data, identify what a good equipment signature looks like and what an early signature for a failure looks like. So we basically give you the capability of collecting data up from five nodes and then analyze them at a later time at your own pace and then create that prediction that is not our core business but that we are trying to enable, adding sensors to the machine to fulfill the predictive maintenance in industry 4.0. Okay, so do I have to build an end-to-end application with ST Win or can I just play with the kit? And Manuel, does ST Micro offer another firmware option for this dev kit? The short answer is yes. We do offer another firmware option and you don't need to implement the whole end-to-end. We provide also a firmware set that will exercise specific functions on the kit itself. And for example, the cool one is to be able to stream all the data from all the sensors at the maximum data rate through USB. So connect the device to USB laptop and get the data out of there directly or even store this data thanks to this iSpeed catalog and store them on a micro SD card. So then basically you can have this kit completely disconnected but still collecting data and then you can go and collect the data at a later time. We try to give you options. You can build the end-to-end. We definitely feel that industry 4.0 end-to-end is where the market will go but if you want to learn what to do with sensor this is a perfectly good starting point that will give you access to all the capability that this device kit has to offer. Excellent. Well, Manuel, I think that's all I have time for today. Thank you so much for joining me. Okay, thank you, Amelia.