Whitepapers

Machine Learning and Artificial Intelligence: The Winning Formula all across the Companies

Maximum degree of automation, more flexibility in production, significantly lower personnel resources: innovative industrial companies hope for a prosperous future through Artificial Intelligence and Machine Learning. And that is not all: PwC‘s consultants have found out that such technologies are an obligatory exercise in staying competitive. The reason: companies make smarter decisions using predictive data analytics and Machine Learning. Also Artificial Intelligence and data analytics are driving the digital factory; “more than half of the companies we surveyed already use smart algorithms to make better operational decisions. Connecting the dots inside the factory and within the company ecosystem, as well as intelligent use of information, will be a “musthave” to stay competitive”, they say.1 However, the final stage of this development is even more intensive in its efficiency: “A fully autonomous digital factory plant which operates independently, based on self-learning algorithms, where people are only required for initial design and setup as well as ongoing monitoring and exception handling. While this can reduce operating costs, main applications include use in hazardous or remote production facilities.”2

 

Yet, the completely autonomous factory is still a dream of the future. There are already numerous examples, showing that companies are on the right track. For example, in the field of robots which is extended by Machine Learning.

 

 

Robotics as a Starting Point

 

Robotics is the perfect starting point for Machine Learning, even before the actual production phase within the smart factory.

 

“Otto”3, for instance, is the name of an inventory movement platform, much more than a self-driving vehicle. This Canadian start-up provides self-driving vehicles designed exclusively for indoor material transport. The vehicles operate with infrastructure-free navigation, offering intelligent, safe, efficient, and reliable transportation within industrial centers. These robots use advanced sensors and Artificial Intelligence to provide flexible automation that does not require fixed infrastructure (no beacons, magnetic tape or predefined laser paths). By added Artificial Intelligence, this solution provides obstacle detection and avoidance and dynamically moves through facilities in the most efficient manner to reach its destination point. The system runs virtually 24/7 because it uses opportunity charging, meaning it autonomously docks and charges itself throughout the day when it does not have a task assigned to it.

 

 

Next Step: Predictive Maintenance

 

Companies and organizations are also preparing themselves for the next steps, not merely concerning logistics or moving goods with the help of Machine Learning technologies. They go beyond probably the most well-known example in this area now is “predictive maintenance”. The popularity of this development is owed to the fact that this concept can also be transferred to traditional factory plants. For example, Dassault Aviation makes use of predictive maintenance for the production of its Falcon 7X business jet, solving three challenges at once4.

 

First, it improves troubleshooting. This is even more difficult, when data on systems’ states and mission history is limited. Second, it helps in the area of testability, because the maturity of onboard diagnostics can be improved. Finally, it improves the company image, because customers these days expect more services from the maintenance system. According to its own statement, the company expects massive profits from predictive maintenance, because the concept applies to up to 30 percent of the aircraft LRUs5. For instance, in the area of hydraulics it can be fluid in pressure and temperature, main and back-up circuits or in the field of electrical distribution, temperature of electronics, batteries and bays, current thresholds, breakers and batteries and differential measurement systems (temperature, current, tension) benefit from this. Given these opportunities, it is no surprise that the predictive maintenance market is booming like never before.

 

According to a new market research report,6 the global predictive maintenance market size is expected to grow from USD 3.0 billion in 2019 to USD 10.7 billion by 2024, at a Compound Annual Growth Rate (CAGR) of 28.8% during the forecast period.

 

 

 

The major factors fuelling the market growth include the increasing use of emerging technologies to gain valuable insights, and a growing need to reduce maintenance cost and downtime. The real-time condition monitoring to assist in taking prompt actions would add value to the predictive maintenance offering and provide opportunities in the predictive maintenance market. Logistics and predictive maintenance are a great way to optimize and improve the existing factories (Brownfield). The next evolutionary step of Industry 4.0 scenarios is that with data science methods such as Machine Learning, individual production processes can be reassessed and transformed. Data is collected and evaluated as par of the production process. This allows individual processes to be better understood and subsequently optimized.

 

One example would be improving sequential processes in the industry, for instance: painting. In most industries, mispainting requires a lot of manual rework To solve this challenge, the best approach would be to record the painting process digitally. Based on training data on paint thickness, pH values and drying times, the process as a whole can often be evaluated and optimized with regard to specific targets such as gap dimensions.

 

Therefore, Machine Learning algorithms bring two major advantages to the production process:

  • improvement of product quality
  • flexibilization of the production process


In this example, data evaluations can lead to processes being continuously adapted to the current production conditions. Smart manufacturing is characterized by the fact that optimizations are carried out automatically and adjustments can be made at the level of individual components.

 

 

Examples from the Pharmaceutical Industry  

 

Even at a much more sophisticated level, sequential production processes can profit from Machine Learning technologies and Artificial Intelligence. This can be seen above all in the pharmaceutical industry7. Companies make use of Artificial Intelligence and Machine Learning today in drug discovery and development. They do so, because pharmaceutical companies are routinely faced with drug development timelines of about 15 years, costs in excess of $1 billion, and a minute rate of success. It is estimated that 1 ut of 10 small molecule projects become candidates for clinical trials and only about 1 out of 10 of those compounds will then pass successfully through clinical trials. Many disease conditions cannot be successfully addressed through the traditional drug development process. And barriers include the inability to devise a molecule that selectively drugs the desired target or the absence of sufficient financial incentives based on the size of an addressable market.  For instance, “twoXAR”8 is an Artificial Intelligence-driven drug discovery company. The company leverages its computational platform to identify promising drug candidates, de-risks the opportunities through preclinical studies, and progresses drug candidates to the clinic through industry and investor partnerships. So what they actually do is to identify drug candidates by uncovering novel disease biology hypotheses supported by real world data. They do “AI Drug Discovery”, by using Artificial Intelligence to screen compound libraries for efficacy against a disease to discover new drug candidates from a public library and identify biological targets. Their researchers analysed the Drug Bank and the Therapeutic Target Database and identified 50 high-probability candidates based on predicted therapeutic potential. This was followed by algorithm evaluation and candidate due diligence to select 10 optimal and novel candidates. They finally tested them in an in vivo model of rheumatoid arthritis to identify 3 lead candidates, all within a 4-month period. Furthermore, they developed a computational model of rheumatoid arthritis (RA) that significantly enriched FDA-approved treatments for RA among the top-ranked candidates. They collaborated with the Asian Liver Center at Stanford to discover TXR-311, an experimental drug for liver cancer. This drug showed positive results in cell-based assays.  Another example is the start-up Atomwise. It has designed a deep learning software tool called AtomNet, which analyses molecular structures for their potential medical uses9. The generative model processes large data volumes and uses rules and simulations (written by chemists) to learn how to generate plausible data. It can also be used to determine how various molecules react and bond to each other. This has allowed research costs for the development of new medicinal products to be drastically reduced.

 

What these companies are doing is that they are bridging the gaps between information and operations. They embed collaborative capabilities between humans and machines – or software-based technologies – to amplify the impact of critical information. Coming back to the OTTO example, as much as 40,000 collaborative robots are expected to be in use in 2020. The annual investment volume on the company side will then be around 180 billion dollars. This is a generation of robots that thinks for itself and performs complex tasks autonomously. While human employees are freed from the burden of repetitive tasks, the company saves resources. According to a research from a German Federal Sate Bank10, sales of autonomous robots will increase by 30 percent annually until 2024.

  

 

According to the analysts, this is also largely due to the fact that in-house development and production control systems as well as the digital connection of the products used by the customer form a fertile ground for the increasing use of Artificial Intelligence in robotics.  

 

 

Becoming a real Industry 4.0 organization is Challenging 

 

Aside from these successes, there are still some challenges, organizations must face in order to become a “real” Industry 4.0 organization, that has reached the full potential of Machine Learning-technologies.

 

First of all, a recent sample study shows11 that companies obviously need help with the selection of Machine Learning models and data. More than 30 percent of the companies in this sector have the largest problems within the framework of Machine Learning projects. This applies to the selection, correction and the „understanding“ of information. 36 percent of companies need better support for the selection of information assets that can be accessed with the help of Machine Learning. For 32 percent of the respondents, help is needed to clean up the data and the provision of background knowledge for the departments („understanding of data“). Smaller companies with less than 1,000 employees require support in the selection of suitable Machine Learning models (39 percent). For larger companies, the percentage is 31 percent. More than a quarter of the respondents (27 percent) need help from external experts to develop use cases in this area. This is especially true for larger companies (31 percent), less so for companies with fewer than 1,000 employees (22 percent). 

 

 

 

 

But these challenges may not be used as an argument to not make use of Machine Learning or Artificial Intelligence.  As McKinsey says: “Leveraging, and transitioning from, digital to new frontier technologies is an imperative”12. But the problem is that while “Artificial Intelligence is on the minds of every executive, these technologies are still nascent”. According to McKinsey: “In 2018, only 12 percent of companies appear to have invested in Artificial Intelligence in every functional domain where the business case for deploying Artificial Intelligence appears to be very strong. The share appears to range from a low 7 percent among healthcare companies to a high of 18 percent in high tech.”

 

The essence of the McKinsey research is that Artificial Intelligence is a new, higherperformance type of digital technology that may boost the ability of firms to accelerate their digital performance. In professional services and retail, companies that do not deploy Artificial Intelligence are reporting digital cash flows that are 15 to 20 percent lower than their Artificial Intelligence embracing peers. In financial services, the gap is 30 percent and in high tech, a very substantial 80 percent.

 

These results from McKinsey are great in all respects, but it would be a big mistake to lump all this different companies together. Surely, larger organizations are not immune to making wrong decisions, but they have much larger human and financial resources as opposed to small and medium-sized companies. Many of them do not have enough resources to embark on the Artificial Intelligence adventure. They need dedicated products and services to be successful to let them remain competitive, even against bigger players.  The message to small and medium-sized enterprises (SMEs) therefore is: they should prepare clear roadmaps when it comes to finding innovative solutions based on Artificial Intelligence. The German Federal Ministry of Economics has found out13 that for instance cloud-based „AI-as-a-Service“ offers in particular will play a major role in medium-sized businesses. “The background to this is that the SMEs are often lacking the corresponding specialists or that their own database is too small to develop their Artificial Intelligence solutions.” Therefore, especially SMEs should not emulate big players, but find out their own way, like some of the examples mentioned in this report. The goal is to allow the SMEs to have a realistic chance to embark on the Artificial Intelligence adventure without large investments, but combining product design and service with Artificial Intelligence. That is the winning formula all across the companies. 

 

The Digital Cashflow Gap:

 

 

 

References:

 

1.       https://www.pwc.de/de/digitale-transformation/digital-factories-2020-shaping-thefuture-of-manufacturing.pdf
2.       https://ottomotors.com/platform
3.       https://blogs.3ds.com/exalead/2019/05/20/industrial-applications-of-artificial-intelligence-and-machine-learning-part-2-of-7/

4.       https://en.wikipedia.org/wiki/Line-replaceable_unit

5.       https://www.marketsandmarkets.com/pdfdownloadNew.asp?id=8656856

6.       https://www.prescouter.com/inquiry/applications-of-artificial-intelligence-in-drug-discovery-and-development/

7.       http://www.twoxar.com/

8.       https://assets.kpmg/content/dam/kpmg/xx/pdf/2018/09/rethinking-the-value-chain.pdf

9.      https://www.lbbw.de/public/research/blickpunkt/20180417_lbbw_research_blickpunkt_ki_steigert_das_industrie40-wachstum_nochmals_7yymebj6k_m.pdf

10.   https://www.lufthansa-industry-solutions.com/fileadmin/user_upload/dokumente/downloadbereich/IDG-Machine-Learning-Studie-2019-de-lhind.pdf

11.   https://www.mckinsey.com/business-functions/digital-mckinsey/our-insights/twenty-fiveyears-of-digitization-ten-insights-into-how-to-play-it-right

12.   https://www.mittelstand-digital.de/MD/Redaktion/DE/Publikationen/kuenstliche-intelligenz-im-mittelstand.pdf?__blob=publicationFile&v=5