XSelect a record to create:
IIC x IoT ONE Industrial IoT Spotlight Podcast EP034: How to sell IIoT solutions using testbeds – An Interview with Mitch Tseng of Huawei
Published on 06/11/2018 | Strategy
IIC x IoT ONE Industrial IoT Spotlight Podcast EP034: How to sell IIoT solutions using testbeds – An Interview with Mitch Tseng of Huawei
*This episode of the Industrial IoT Podcast is sponsored by the Industrial Internet Consortium
How to think about building an IIoT solution to provide business value? How to communicate IIoT solutions to corporations who want to implement some “IoT stuff”? What is the role of organizations and associations in accelerating adoption of IIoT solutions?
Mitch gives us an overview of the manufacturing quality management testbed at IIC in conjunction with Huawei and Haier. He also gives advice on the process of conceptualizing an IoT product, and how to best go to market with it.
Mitch is a managing member of Tseng InfoServ, and a distinguished consultant to Huawei at the IIC. He is also the co-chair of the IIC Innovation and Edge Computing task groups. https://www.linkedin.com/in/mitchtseng/
The Industrial Internet Consortium is a global, member supported, organization that promotes the accelerated growth of the Industrial Internet of Things by coordinating ecosystem initiatives to securely connect, control and integrate assets and systems of assets with people, processes and data using common architectures, interoperability and open standards to deliver transformational business and societal outcomes across industries and public infrastructure. http://www.iiconsortium.org
IoT ONE is an online platform devoted to accelerating adoption of Industrial Internet solutions. We are mapping the global ecosystem of IoT vendors, use cases, case studies, and technologies. We leverage this data to help companies source technology, research competitors, and enter new markets. https://www.iotone.com
Industrial Internet Consortium Manufacturing Quality Management: https://www.iiconsortium.org/manufacturing-quality-management.htm
Industrial Internet Consortium Member Networking: https://www.iiconsortium.org/iic-connect.htm
Industrial Internet Reference Architecture: https://www.iiconsortium.org/IIRA.htm
Welcome back to the Industrial IoT spotlight. I am joined today by Mitch Tseng. Mitch is a managing member of Tseng InfoServ and a distinguished consultant to Huawei at the Industrial Internet Consortium (IIC). At the IIC, Mitch is the co-chair of the Innovation Task Group and also the Edge Computing Task Group. Mitch also has a deep background in some of the fundamental technologies behind the Industrial Internet -- he's involved with the International Organization for Standardization and a chair at TIA engineering committee for vehicle telematics. Mitch is the perfect person to be speaking with today about how we are taking some of the fundamental technologies that have been developed in the past decade, and now putting them into very practical use cases that traditional businesses who don't have deep I.T. departments can start to implement. Mitch, thank you so much for joining us today.
Erik thank you for having me.
Mitch you've got an interesting background. You also have a Ph.D. so you are very much a technologist. How did you end up you know first setting up Tseng InfoServ and then representing Huawei at the IIC?
Several years back. I was focused on the telecommunications are many years. My Ph.D. was working on digital signal processing and then eventually getting into wireless communications. After working in the industry with those major players like Nortel Networks and Nokia, I found that the world is changing and all the communication bases back then form a very good foundation for future work. So I left Nokia in 2009, I decided I’m going to look for the next step, and I found a new love – machine to machine communications (M2M). That was the time when people were talking about small machines and the system on chips and then eventually try to get to using sensors and networks to help our industry and to boost our life quality. And then later on, of course, M2M became Internet of Things (IoT). So I started looking into that part and tried to leverage my background in communications. I am also very interested in the business models – there was a time actually I was assigned to Beijing by Nokia, focusing on smartphone platform marketing. So I've been trying to combine them together, and I realized that IoT is not something only for the technical people, but the success is actually a combination of the technology and the business.
So I formed Tseng InfoServ as a consulting firm, started talking to people, and eventually became an IoT evangelist. In the process, I received attention from many organizations and also some companies and eventually I ended up with a contract with Huawei Technologies.
Huawei is really one of the more interesting companies out in the market. A lot of people know them from their consumer business, but they are a leader the enterprise market in the connectivity infrastructure space. Are they now the number one player in the market globally?
Arguably yes. I think their total sales in the infrastructure side surpass all competitors to become the number one in the world. Of course their fundamental business is actually access – radio access and wireless access – and eventually they branched into enterprise networks. There are two parts – the terminal handset and the network. I've been working in the terminal research side and also now with networks.
The testbed we're going to be discussing today is manufacturing quality management. This is a very operational side of a business; it's identifying quality issues in manufacturing. So it’s interesting that a connectivity company like Huawei would be taking the lead.
What's the background story? Why did Huawei become interested in this topic and decide among all the potential topics that could be the focus of a testbed, to focus on manufacturing quality management?
In the IoT or IIoT industry, it is actually a combination of the IT and OT people. Huawei has a very strong presence and experience in connecting terminals and building networks, so they just try to support the whole communication community with the best access technology.
They are very strong on the technology side, but on the other hand, they recognize that the pace of growth of the communications business with traditional telco operators is slowing down, so they need to figure out what is the best way to revive their technologies for the future. When the IoT industry becomes more practical, there will, of course, be the commercial side with wearables, but there will also be an operational technology domain which no one in the IT community has adequately addressed. For example, a lot of factories have automation, but they are using human controllers to make sure that the process is running well. The objective for these factories is to make sure that the production line produces good quality and quantity. In that case, they are very content to keep the status quo as long as the current production line is not breaking down, they can expect business to run as usual. But now, with the advance of IIoT, the IT people realize that there are a lot of things that can help to improve the production line. For example, sensors are getting cheaper with the quality, computing power, and memory size growing exponentially over the past few years. So we can probably improve the conventional production line with our new technologies and devices. However, we need to figure out a way to introduce them and get them to accept what we can offer, so that we can realize all the potential benefits that these new technologies can bring, and create a much better future.
There is Industrie 4.0 in Germany, and Made in China 2025 in China. Both are based on joining the IT and OT together to produce a much better product. The process could also be streamlined to become more efficient, and could even be fully automated. The whole industry is looking at this. Huawei, being the leader in the telecommunications domain, figured that they also need to get involved in this area and that’s how they joined the IIC.
It makes a lot of sense in the context of China's position in the world as a manufacturing hub, that this manufacturing quality management would be an intelligent place for Huawei to focus, as they move more into the the OT domain.
A couple of the other stakeholders in this test bed are Haier group, one of the larger white goods manufacturers in the world, China Academy for Information and Communications Technology, and China Telecom. One of the strategic priorities as I understand in China right now is to improve the quality of manufacturing. China is very efficient on quantity and cost, quality is already excellent in some cases but there is still a lot of room for improvement in other cases. So it makes sense to focus on that as an objective for Huawei and the other stakeholders. What I see as the goal on the testbed website is using the MQM, the manufacturing quality management analytics, engine to drop the false detection rate by 95 percent. So let's get a little bit into the business case here. For Haier, the end user, what's the business case behind this for Haier or for another manufacturer that might be using this solution in the future?
The original idea for this MQM testbed is to find a systematic way to renovate existing manufacturing facilities in China so that they can fit into the future production system. Huawei and Haier have teamed up because Haier focuses on electronic appliance manufacturing, so it’s the perfect OT partner choice. When we first started, we heard there was some problems with the welding section, that was a key component of the air conditioners that Haier was building. They have some problems in false detection because faulty machines were passing quality tests and being shipped out to customers. This causes a huge cost because the customer wants to return it and Haier needs to process this return.
So we got together with all the members in the testbed and try to help them to control the manufacturing quality. As with an engineering process, we tried, at first, to focus on the welding section to zoom in on the root problem straight away. We found that it was not all that straightforward, and we then looked at the quality control (QC) stations. We managed to improve the pass pass rate from 50% to 95-99%, which means that almost everything that passes is supposed to pass.
As you’re walking through this, it would also be great to understand the technologies involved and how to cost efficiently integrate new technologies into a brownfield factory.
At the beginning, we jumped into the welding section and try to detect the problem using the IIC structure. As we know, the IIC 3 tier structure, contains a lot of sensors and sensing networks passing data to the second tier with the analytics, and finally the third tier of management to ensure that the process runs smoothly. When we applied this to the welding section, we realised that the welding environment is a very “dirty” environment electronically. There are high voltage arcs running around, and the hardware has been there for some time. So after a few trials, we recognize that it is almost impossible for us to stick a new sensor in there just by retrofitting. Some IIC members who have welding facilities actually reached out to help, like Fujifilm and Olympus. We found out from them that most of the time, they do not do a real time chip; they do post analysis. For example, if you produce mobile phones, they will take a x-ray of this phone post-production to check if there are any cracks. The technology is based on ultrasounds; they run the whole material through the ultrasound system and then based on the response, they can detect the cracks. So we know that there is no real time process to detect the cracks. So we pulled back and looked at the whole quality control system, and not just one station.
On the second visit to the factory, we learnt that the QC is done using listening – they would have 2 or 3 experienced listeners who turned on the machine to listen to the annoyance of the machine. As the process is highly subjective, it is not a standardized method of detecting failure. So we focused on the analytic engine part, the platform side, to see if we can get something that we can use to help this acoustic noise detection process.
I have some expertise on the noise detection process before. We can do something with some kind of training data that can result in an automatic process. Fortunately, Huawei had an artificial intelligence (AI) engine at that time. We borrowed it and used a lot of training data from Haier to build a deep machine learning algorithm. We found that the machine actually picked it up and processed data in a relatively reasonable fashion so that we did not have too much of false positives.
When we first put it to the field test, the machine could at least maintain the same level as a human listener. We wanted to improve it to do better than humans. We analyzed it and realized that there was a lot of ambient noise, so we built a noise-proof chamber for the machine to operate in. The first file we ran in this environment reached almost 100% accuracy.
At that time, the higher management was sceptical about what we were doing because a lot of people have been talking to them about “this IoT stuff”, but they never had a good feeling about how this stuff can help them. What we did was to use the IIC architecture and get the management to agree with us so that we can operate on their production line. Eventually when we showed them the great results, they were fully committed and actually approved a second project on another product line.
That's a great point that you made at the end of this kind of process brief.
I was in a conversation yesterday with a partner and you know she mentioned that she's working with a lot of Chinese manufacturers who say you know they want to know which technology they can use to improve their facilities and they want somebody to tell them install these or purchase these set of hardware and software and you'll become more profitable and “an industry 4.0 ready” facility. That's just not how it works. The process that you just outlined here very much a learning process where your initial hypotheses didn't really pan out because you found out that the operating environment was too noisy and uncontrolled in order to install new sensors to accurately collect data and have a result there. And then you tried to deploy the analytics engine or the algorithm in the existing situation that the human is operating in, and found out that there's too much ambient noise for this to be an efficient solution.
So is this iterative process where you change scope and you had to probably understand the problem and then develop this solution around that. That’s just where we are, so it needs to be kind of a learning process – a discovery of where in the process can we actually use technology to create a result. And then, which technology suite might that be and it might be a completely different set of technologies or different approach than we initially envisioned.
Where are you in that testbed today? It sounds like you've had a great success; Haier is this open to expanding the scope. Is this now a solution where you say it's more or less ready to to bring to market? Or are you working with partners to deploy in different environments and continue to refine the approach of retrofitting a brownfield environment? Is it already standardized enough that it's commercial ready?
The process is now at the point of wrapping up our findings and publishing the final report on this MQM testbed. We are now seeking IIC members who are interested in applying this methodology to their process. In the process of the review session, one of the causes of why IoT adoption has been slow is maybe because we should be trying to work with a client and understand the problem first, instead of building products that sound great but does not fit client requirements. For example, a sales person can go to a client and show a great invention, but it is not ready for client usage. We can do better by listening to the client first, defining the problem, then coming up with a solution to create immediate tangible value for the client. Right now, we are using the IIC as a platform to broadcast that, and encourage more members who are interested in doing the whole process to discuss with us via the IIC.
If I can just kind of redirect or synthesize that, I would say if you're an industrial IoT technology company, emphasis that needs to be placed from the beginning on building partnerships. That's one of the goals of the IIC testbed program but typically these companies have a great product that fits into a much larger solution that is going to require a lot of customization. That means that they need to figure out, how does my product fit into this solution and how does this solution have to be customized? As you said, typically the technology company has some engineers sitting somewhere in the world where they're very disconnected from actual manufacturers, trying to figuring out how they're going to get to market, how they're going to make their specific technology fit into the system and configure that system for the needs of real client. It's not going to be something that most companies are capable of doing by themselves, certainly the Siemens' and the GE's can potentially do that, although I think they also are seeing more value in partners. But for most companies they're going to need to get some system integrators involved, they're going to need to get some other companies that maybe sensor manufacturers and so forth that have really unique vertical expertise and collaborate with them on getting that product out to market. I think companies are starting to take more of this ecosystem approach; it's just necessary because even a large company like Huawei is providing one set of technologies into a much broader system at the end of the day. They don't have the vertical expertise, so they certainly need to collaborate together. That's a big strength of the IIC testbed program.
Let's turn to more of a deep dive on the technology. So when I'm reading through the end of the MQM brief here, I see it broken down into the cognitive computing platform that Huawei has developed into data acquisition, data preparation, cognitive analysis, and testing. Can you give us kind of breakdown with the actual technology suite looks like? What were the sets of technologies that were developed during this process? What are the technologies that were kind of already off the shelf either at Huawei or partners that you integrated? What was needed to be developed specifically for this solution and which technologies do you use that were readily available on the market?
Some software was off the shelf, some were developed by the Huawei AI research lab. The deep machine learning engines were developed internally.
What about the engagement of China Telecom? I see they’re one of their member participants. Since connectivity solutions are required here, was it more or less a standard connectivity solution that they provided or did they also have some strategic or R&D support in this testbed?
In our original design, China Telecom will be the provider if we want to do the remote connectivity.
For example the AI engine is in Haier’s factory – that’s what we want to do for Phase 1. For Phase 2 and 3, we want to beef up this network in the factory so that we can provide a better connection to make sure that all data can be connected, even beyond this quality management session. In our proposal, there will be a centre in Beijing area, and they will have a mirror system, which means that whatever happens in the factory will happen in the mirror onsite. Hence, they can do all the heavy reconfiguration or software upgrades through the network in another city. These connections will be provided by China Telecom and CEICT.
Interesting. As you continue to identify the right solution for this problem, was it the case that that was just not necessary for the solution that you developed or is that kind of a hypothetical next stage? It sounds almost like kind of a digital twin type solution if I'm capturing the concept right.
The original proposal was to finish it as a baseline project in two years, with the main objective of going to the field to understand the customer’s needs and to come up with a workable solution with a tangible result. The decision at that point is that, “yes the AI machine works, and maybe we should just go to work on that”. But in the initial phase, that may be too cumbersome. So what we are doing is shifting the whole computer to the production line. Some people may think it is overkill if you just want to do this acoustic detection because of all the computing power dedicated to it. But for us, this is more than a proof of concept, because we also use this to demonstrate to the OT partners that this is what we meant by an analytics engine, and they can envision using that in their own factories. The digital twin concept is in the pipeline.
Clear. I suppose what you've solved so far is you've spelled out one specific problem domain, and then this digital twin concept could potentially monitor the entire factory where there might be 100 kind of similar deployments that would be monitored at different stages in the process and could be adjusted and optimized over time. So I could see the value especially for a larger organization to have kind of a central control center where they're able to monitor the different production lines, and different stages on those production lines, remotely that certainly is much more cost effective than having troubleshooting teams in every facility because there's just so much talent to go around right.
Exactly. We heard from our partners that the management was so impressed with this project that they are forming internal teams to focus on how to retro fit the rest of their production lines with this concept.
What do you think is the next step for you? Are you going to be taking this through to the next stage, or do you see another potentially another testbed that might be on the horizon for Huawei?
Huawei has multiple testbed in the queue right now. To name a few, a digital metro subway system, an elevator network connecting all elevators for advertising, surveillance, maintenance, etc. There are a lot of people in Huawei who are looking as testbeds as a way to showcase what they have and to find partners for the future. This way, they can turn these ideas to form a connected world with much better access technologies. Personally, I will just try to be the spokesperson for MQM, and connect with interested parties and lead some other testbeds in the future.
As a closing thought based on your experience with MQM, what advice would you give to companies that are either an IIC member and haven't gotten involved in testbed or not an IIC member but a company that's trying to figure out how to do cost effective development of industrial IoT solutions? Whether it's joining the IIC or just figuring out how to run programs with multiple other stakeholders with multiple partners? A lot of companies just haven't had much experience with that in the past -- it's either been internal or it's been bilateral. This model of having multiple stakeholders is fairly unique, and not particularly easy, but very impactful and successful. So what advice would you give to companies that are exploring this concept in order to help them maybe take the first steps and minimize the risk of failure?
This is a million-dollar question. I've been working on this M2M and IoT domains since 2009. It never surprises me that there's so many people interested in it and new products everyday. There are some companies who think that they can turn rich overnight just because they have a new product, but the truth is IoT or IIoT is actually an end to end business. You cannot just say “OK I've become part of that and become very successful”. That means when you try to think about what you want to do, you have to consider how this piece of device will fit into a much bigger system.
If you have a sensor then you need to know how much information you can get, how much power you consume, and how secure it is because all these characteristics will affect the way it goes into service. When we talk about IoT, we actually talk about IoT service. If you're thinking about IoT without think of IoT service, you're going to run into trouble in the future.
What I learnt from the IIC is that you can test whatever you want, but you have to make sure that you have an end to end service, because without it, you cannot provide value to your customer. The IIC process forces you to consider the edge along with the data analytics as well as the management side. This is an end-to-end process. We have now 28 testbeds in the IIC and I can confidently say that you can just put a business model on top of each one, and it will become something that you can try to run as a business, because they are all end-to-end services.
We have a system in China for water monitoring. It works with some county officials to monitor if there is a leak in the pipe by comparing the water consumption in households with the amount that is pumped out from reservoirs. This business started from a testbed.
For all the people who want to be successful in IoT or IIoT, firstly, you have to really think about the end to end, and where you want to fit in in this process. Secondly, you have to think about how and who to partner with because you’re only one part of the process. There are a few ways to do this, either by joining an association or going to a platform where suppliers showcase their capabilities and finding one that suits your needs. In the IIC, you can connect with people by registering and posting your free time slots so that people can call you for a discussion about what you want to jointly do together. I believe that the IoT business can be proliferated soon because volume matters. If more people engage with IoT, deployment will accelerate because everything will be more easily accepted and efficient to implement – it is like a positive cycle. If I can be of any help, you can reach me on LinkedIn or email email@example.com