Industry 4.0 requires large scale, global machine data collection.
EMS providers are moving toward Industry 4.0, utilizing complex machine learning (ML) and artificial intelligence (AI) to improve productivity and control costs. These new tools require massive amounts of data to perform increasingly complex data analysis that improves manufacturing decisions. The existing ecosystem of small-scale, tactical data collection systems that provide intelligence about real-time factory operations lacks the ability to collect historical machine data at scale.
Differentiating tactical and strategic data collection
Clearly, an EMS factory cannot function without the benefit of tactical data. This is the data required by line and building supervisors to effectively manage day-to-day operations. Rapid responses to everything from a warning light on a piece of equipment to a rapid rise in plant temperature keep the flow of production on track. Doing so while keeping costs down means fewer technicians managing more machines.
This widespread need for collection of tactical data resulted in a wide variety of commercial systems that provide operational data. Platforms able to collect strategic data — the historical data that can be aggregated and analyzed long after a shift or production run ends — are few and far between.
Historically, the cost of collecting and storing so much data was cost prohibitive for most EMS factories and, until recently, the data analysis tools capable of pulling insights from large historical datasets did not exist. Many EMS factories in the past generated but did not store the vast amounts of transient operational data used to keep production flowing.
Establishing the value of strategic machine data
As the role of AI and ML in Industry 4.0 comes into clearer focus, EMS providers should look beyond the tangible and immediate benefit created by tactical data and search for more data and deeper insights locked inside factory machines.
For example, pick-and-place (PNP) machines pull electronic components at high speed from component feeders and place them onto circuit boards during the printed circuit board assembly process. PNP machines often provide a real time fault alert when the error rate of picking parts from a single feeder exceeds a defined threshold such as: If three parts in a row from a single feeder are lost, the machine will stop and alert a technician to inspect it.
But what if that machine consistently lost two parts in a row, never meeting the three-part alert threshold? Imagine it does this day after day, week after week. One machine creates component attrition every minute it is in operation, but the problem is never acute enough to trigger a tactical alert.
This is the type of problem that strategic data analysis addresses. It looks at collected data over long periods of time to identify trends that amount to acute, slow motion tactical problems, and creates the ability to resolve such issues in minutes versus weeks or months.
Assessing strategic data collection challenges
So what capability does a system need for EMS factories to be able collect and store the historical data necessary to fuel ML and AI capability? A platform should collect rich, contextualized data that can be unambiguously interpreted when combined into a large historical dataset. It also needs to combine data from as many machines, lines, and factories as possible for analysis that compares a broader array of observations — offering more behavioral examples that can support deeper analysis and more powerful conclusions.
But centralized strategic data collection is a different animal from operational data collection systems that are typically small-scale and often disconnected from each other. The requirements for collecting large volumes of centralized, strategic machine data present several unique challenges:
- Systems connected to all these lines at once must account for reporting reliability, the type of analysis being performed, and the stability of the analytic tools in use.
- Integration with other systems, specifically manufacturing execution systems (MES), requires passive connections to each machine to avoid potential shutdowns in the event the data collection system becomes unavailable.
- It’s critical to move beyond common misconceptions about the volume of data created by centralized collection systems. Modern, cloud-based data management software is designed to handle petabytes of data without issue. The data volumes generated inside of EMS factories, even large ones, do not exceed these capabilities.
- Currently, there does not exist a standardized or widely established method by which surface mount technology (SMT) machines report performance and error data. Machines from different manufacturers and eras will produce and report data using different formats.
- Experience has shown that for centralized data collection systems, avoiding any kind of manual data entry step is best whenever possible. Manual data entry or enrichment is slow, cumbersome, and an invitation for error.
Overcoming obstacles to large scale machine data collection
More and more EMS providers are coming to the realization that large amounts of machine data generated during production is retrievable, usable, and valuable. They are accelerating production, reducing waste, and lowering maintenance costs. To keep pace, you will need domain-specific solutions that collect machine data and produce a steady stream of alerts and actionable information.
Arch offers a technology solution that reliably extracts data from any and every machine on the production floor; from the new ones with modern APIs and the ancient ones with data trapped inside. Arch has developed a standard library of machine connectors that includes vendor application programming interfaces (APIs), numerous types of readers, and retrofit sensors. To learn more about our real-world experience designing and implementing a global factory data collection system across a large network of EMS factories, read our technical paper detailing the process.