As companies progress in their smart manufacturing journey, one sure outcome is the generation of huge amounts of data. Whether they can harness this data to drive better decision making is an important aspect of success. This can be done by understanding data characteristics and following simple practices.
There is a massive variation in data required by individuals in different roles for decision making, as well as the impact of those decisions.
Rockwell’s Annual State of Smart Manufacturing Report 2023 found that “access to useful data to make effective decisions in real-time” is a top five leadership obstacle for manufacturers. Thus, system design and implementation teams must carefully consider many aspects of data to ensure the right data is delivered to the right people at the right time and in the right format. Key things to consider include:
Data context: A data value in itself is only meaningful if it’s contextualized.
Weight dimension: This signifies how much processing has been done on a piece of data. Contextualization of data also contributes to its weight. The higher the weight, the higher the place in the data, information, knowledge, wisdom (DIKW) pyramid. For example, the instantaneous temperature of a bearing is “data;” a trend of such temperatures is “information;” analysis of temperature variations based on operating parameters is “knowledge;” and patterns that lead to bearing failure are “wisdom.”
Time dimensions and consumption lead time: Lead time between data generation and consumption can be anywhere from microseconds to days. For example, data is consumed by a process controller in microseconds, operators monitor the same data in seconds and a user receives a notification related to this data in minutes.
System configuration should ensure the shortest lead time for any given scenario. This increases the agility of decision making. For example, if the hourly rate of production is 20% lower than the standard rate, notification should reach a supervisor within a few minutes so that it can be corrected. And a failure prediction alert should reach maintenance within seconds so they can plan actions before the failure occurs.
Consumption lead time and weight also correlate with who consumes the data. Operational people normally consume lightweight (raw) data. Plant management uses heavyweight data like plant efficiency trends with a longer lead time.
Time period: The system configuration has to ensure an optimum time period and processing logic to allow for meaningful decision making. For example, an average takt time value for one week is not meaningful if multiple products were made. However, an average takt time for one product is useful in decision making.
Another example is a machine-learning (ML) application’s use of data for predictive maintenance. In this case, the longer the time period, the more accurate the predictions will be.
Useful age/longevity: How long does data remain useful after generation? This is an interesting aspect and gives insights into how long data should be preserved.
Data longevity depends on its weight as well as the context. For example, the process parameters of a silo are critical for real-time monitoring but don’t have much value after the batch is complete. Longevity of the same data will be higher if the process was abnormal or if a batch produced turns out to be a “golden batch.”
Velocity: This is the rate of generation/useful consumption of data. Velocity is inversely proportional to weight and longevity. For example, the velocity of vibration data may be 10 values per millisecond when an ML algorithm consumes it. The velocity of plant efficiency data is one value per day. Even though the system can calculate efficiency every minute, its useful consumption cannot be done at this velocity.
Before Industry 4.0, industrial data flows were hierarchical as defined by ISA 95, and the amount of data flow from plant to enterprise was small. With smart manufacturing, these data flows have exploded and become more complex. To manage this complex data environment, using the following practices is imperative.
Standardize and contextualize data: This allows data to be consumed more effectively by users and analytics engines.
Minimize data delivery: This may seem counterintuitive, but delivering only data that can be used meaningfully is important. It’s also necessary to fine-tune data velocity to the correct level.
Reduce consumption lead time: Achieving this, critical data should be pushed to users via notifications. Users can pull the rest from the system as needed.
Manage data flows across systems: Identifying and controlling data flow across systems should be possible. This is important from a security standpoint as well as for change management. Industrial data-ops frameworks can be used for this.
Design data disposal strategy: Keep the system lean and healthy by ensuring outdated data is not lying around. Consider all different types of data, as well as your full range of data design mechanisms, when deciding what is outdated and needs to be cleaned up.
The greatest benefits of IT-OT convergence and the biggest paradigm shift of smart manufacturing stem from using real-time, accurate and consistent manufacturing data.
Using the considerations outlined here, companies can climb the pyramid of data, information, knowledge and wisdom to push their performance to a new level.
Connect With Us
