Issue link: https://iconnect007.uberflip.com/i/1216587
44 SMT007 MAGAZINE I MARCH 2020 your own system or select the right solution providers and partners. Also, when selecting a new architecture, make sure it allows you to scale the amount of data you can collect with- out paying higher marginal costs or sacrificing system performance. At level three, your data is active. A level-three system shifts manufacturing operations from reactive problem-solving to proactive analysis and improvements. The system enables opera- tors and engineers to be truly preventative and proactive in solving problems. To move from level two to level three, new system capabilities—such as machine learn- ing and AI—are added to the previous level's data architecture. These new tools allow you to start generating insights in as little as two or three months, depending on your product mix. Built onto the level-two system that aggregates all your production data, these new features create an intelligent system that finds valuable insights and predicts failures more accurately on its own while delivering information to the appropriate person at the right time. Engineers do not have to query the system or perform manual process analysis to find the answers to solving production issues. An example of level-three system attributes includes machine-learning models that predict product defects or machine failures and iden- tify ways to produce products more efficiently. In a level-three system, a person is still needed to make the changes that the intelligent system recommends. Level four is when your data becomes action- oriented. At level four, the data system deploys the recommendations that it finds from ana- lyzing manufacturing data. For example, a machine-learning model will identify an opti- mization, then generate and send the recom- mended new settings to the machine where it is automatically executed. In such a closed- loop, AI-controlled production line, the time it takes to execute on an insight discovered by the system becomes minimal. Achieving level four requires datasets that are large enough and have enough validated cases to provide the information needed for the system to "know" the effects of a produc- tion change. The time needed to move from level three to level four varies based on the amount of time it takes to gather the necessary datasets. Looking at building a smart factory in these four stages is helpful when making such a fun- damental, and even monumental, change. No shortcuts can take a manufacturer from level one to level four. Those that have tried find their systems have so much process and data variability that they quickly become mired in complexity; they built on an unsound founda- tion with weak construction. Figure 2: Digitalizing the entire manufacturing process.