The invention of the combustion engine was a significant technology innovation, but its transformative capacity was unleashed only by process innovation—the power of the assembly line. Together, technology plus process combined to revolutionize the transportation industry.
Similarly, the availability of immense amounts of data from sensors throughout the manufacturing environment is a major technological phenomenon. But without process innovation, that data is just so many bits and bytes.
As documented in our recent white paper, “Driving Improvement in Manufacturing through Advanced Data Analytics,” we’re metamorphosing that data into wisdom—and generating considerable business value—by creating processes that support real-time decision making.
In my role as Manufacturing IT Senior Principal Engineer for Intel IT, I have found that a crucial component of those new processes is the data life cycle. Thoroughly understanding the data life cycle and managing it more efficiently allows Intel IT to protect information, solve complex problems, and manage the IT environment effectively.
Data Life Cycle is Key to Harvesting Wisdom from Data
Each stage of the data life cycle includes activities and requirements to optimize the value of the data. The data life cycle consists of the following stages:
- Data storage. We use storage methods that can handle hundreds of data points per second, terabytes of summary data, rapid read/write capabilities, and archiving. We’ve carefully designed our storage system with an eye toward future data growth.
- Data mining. Mining the data in accordance with Intel Privacy Principles (see the Intel white paper, "Applying Privacy Principles in a Rapidly Changing World”) helps discover correlations in data from a single source and reveals new insights. We choose CPUs and networks that can meet high-performance computing needs. We think of it as the following equation:
raw transactions (data) + business logic (information) reports (wisdom)
- Data integration. To solve complex problems and create meaningful insights, we perform data integration across disparate sources. This integration provides multiple, distinct viewpoints that, when combined, reveal a holistic view.
- Notification. To make notifications as effective as possible, we prioritize them, deliver them to appropriate devices, highlight what is most important, and when appropriate, prompt a response.
- Reporting. We deliver actionable information tailored to the device and user receiving it, using HTML5-based content that is clickable and easily consumable.
Manufacturing Analytics in Action
The following two examples illustrate how much business value can be generated by concentrating on the data life cycle, using it to power advanced analytics and targeted reporting. You can read about more examples in the IT@Intel paper, “Joining IoT with Advanced Data Analytics to Improve Manufacturing Results.”
Example #1: Evaluating Manufacturing Tool Health
Outcome: Analysis that used to take 4 hours now takes only 30 seconds
Engineers at Intel’s factories are responsible for tool health. Using sensor data and investigatory interaction, these engineers can now run analyses that determine what sorts of tool health data is meaningful, what decisions need to be made in which timeframes, and under what conditions should notifications be sent.
Example #2: Consuming Large-Scale Sensor Data (aka fault detection control)
Outcome: Ability to process over 5 billion data points per day, resulting in measurable improvement in equipment availability and yield
Many of our factory sensors on equipment collect hundreds of data points per second per sensor. We extract this data and use it for both real-time detection and end-of-line correlation. Engineers and manufacturing technicians use the meaningful results to refine equipment behavior.
Because sensor data can often be large and complex, we consider the following when designing our storage systems and algorithms:
- How will the data be consumed?
- How is the data stored (schema)?
- Where does the data reside? How fast can it be retrieved (affects hardware choices)?
It is important to start with something simple and build from there. The time series analysis we needed to do is quite complex, but our first attempts began with only a single variable. (Remember, Rome wasn’t built in a day.)
Start Simple and Grow the Solution
We have thousands of sensors collecting massive amounts of data points every day; the resulting data sets are huge. Over time, we have allocated resources and tools to help us manage and harvest the value from this data. For those getting started, realize that the potential of data doesn’t require a vast network of sensors in the beginning; it starts with understanding how to manage and optimize whatever sensor data is available.
Intel IT started small and experienced a learning curve in mining and integrating data across multiple sources. We then were able to identify what was most important in the vast sea of information and we soon discovered actionable insights. Our algorithm development started with a single variable, then we slowly added complexity. Now, our analytics activities have grown to support increasing amounts data, in addition to more traditional sources. I’m confident that as others undertake this journey, they will also realize the benefits they can obtain from their own data.
Read our paper for additional information, and then begin to put your data to work. You will be amazed at the results!