The Secret to Manufacturing Intelligence
Over the last year, I have written extensively about why analytics is the basis for enterprises to compete more effectively and as well on why analytical initiatives need to start with the existing set of business capabilities. One area that is sadly in need of better data and analytics is manufacturing. The simple fact is the data remains siloed and disconnected from the rest of the business at organizations that praaoduce products for a living. The reason is that the manufacturing side of these organizations are locked into their legacy architecture of sensors, actuators, controllers, and HMI devices.
Driving efficiency when demand is volatile
At the same time, manufacturing businesses say that they need to drive operational efficiencies and productivity improvements in order to remain competitive. It seems clear that many if not most are facing continuous downward pricing behavior at the same time as they find their demand from customers increasing volatile. Making this especially difficult is the fact that manufacturers have to commit to creating inventory in advance of receiving purchase orders. In semiconductors for example this can be particularly severe with product lead times running 8-12 weeks from wafer start to final production package test. In this model, it is essential—including for fabless vendors—to be able tweak wafer starts by knowing where unfinished goods are in the process and to be able to relate this data to the demand that is predicted for future periods. The same can be said for just about every continuous or discrete manufacturer.
Limited ability to analyze business and manufacturing processes
Getting after these issues typically requires better management of third parties. For most, improving yields can actually start with third party vendors. Clearly, fixing things here can also involve taking steps to improve operational processes and to link them to business process. Historically, there has been limited ability to do this and more importantly, to conduct business level analysis. This has limited both the effectiveness and efficiency of manufacturing operations.
Creating a manufacturing ecosystem view
To manage better, we need to create a total ecosystem view. We need to connect internal and external sources for suppliers all the way through to customers or put differently, from supply chain to distribution chain. For these reasons, data sources range from back office to front office and from suppliers through to production and onto customers. The below chart demonstrates all of the potentially relevant content. Clearly, the most interesting problems to solve here range from quality/yields, cost reduction, and customer demand. And none of these rely on just a single source of data.
The biggest lack of connection, however, is the fact that operational manufacturing data has stayed operational and in many cases relied on tribal knowledge corrections to adjust data coming from out of kilter manufacturing sources. However, three big trends are driving change in how manufacturers look at data—big data, data science and predictive analytics, and streaming real time analysis and workflow. So what needs to be overcome to take advantage of these trends and bring manufacturing into the business fold?
Big data changes how and where manufacturing data is collected
Clearly, what big data enables is the opportunity to go direct to the sensor as a data source. However, in contrast, to other big data opportunities, it is essential that we correct sensor data before we do any analysis. Here, we need to deal with the semi-structured nature of manufacturing data. Without this context, what does the number 84 mean? Is it a temperature? And if a temperature, what kind of temperature is it? At the same time was it created from a defective manufacturing system sensor that has a known drift. Context and drifts need to be addressed before data enters a Hadoop cluster or an enterprise data lake. Clearly, the opportunity is with correct and contextualized data to relate manufacturing data to appropriate business data. From the manufacturing and business perspective, the opportunity starts with the following:
- To connect incompatible legacy manufacturing data sources to the outside world
- To automate the correction of sensor data drifts in advance of automated prescriptive action
- To apply predictive analytics and machine learning to automate actions and to make operational time series data usable by predictive analytical and streaming solutions
What is the big data opportunity for manufacturing?
So if we put it together, what are the big data opportunities? It is to automate the response to streaming manufacturing data according to business rules. Next, it is the ability to create an integrated total business data lake based upon Hadoop and including predictive analytics. And finally, it is to enable a Big Data Based Historian. With these in hand, the next opportunity is to fix manufacturer supplier, product, and customer data. Part of this is to make sure on the supplier and customer side that you have picked the right partners including checking their linked financial business position. At the same, it should be clear that as we link and integrate manufacturing data and correct for data issues, we can apply where relevant machine intelligence. Finally, there is the ability to discovery relationships and correlations. So much of big data today is about simplifying what were complex and difficult integration processes. Today we have the opportunity to discover what data goes with what and then create reporting that surfaces patterns needed to make better decisions. Regardless of which portion you use first, the opportunities here are four: 1) a single view of customer; 2) proactive management of suppliers; 3) proactive management of manufacturing and the business; and 4) better management of financials of the manufacturing businesses. For all of this to work, companies that manufacture products and data need to create timely, trustworthy, and connected data to really impact their decision making processes. Big data alone is not the answer. This means that they need to be able to pull data from supplier manufacturing data, supplier quality data, manufacturing quality data, manufacturing sensor data, ERP/MRP/MEM, and Salesforce automation. But the key enabler whether on it is on vendor side or the end product manufacturer side is the ability for the first time to collect data direct from the sensor without going through the logical controller, SCADA/HMI, Historian, or Manufacturing Intelligence layers. Vendors that take this step and make the data created shareable to their customers will strengthen their competitive advantage in an increasingly competitive world. In the end, all of this makes it possible to correct data on the fly and then within a big data repository relate this sensor source data and to make decisions from a manufacturing or business prospective. This includes on driving better third party decisions, achieving required product manufacturing yields, and driving more efficient operations as a whole. Doing better here is not only about more effective and efficient operations. It is about as well detecting early product problems—hopefully before they reach the customer.
Manufacturers today needs to drive operational efficiencies to drive the productivity improvements in order to stay competitive and relevant. This includes having the ability tweak manufacturing volumes to minimize uncommitted inventory volume. With this said, data is increasingly a business capability that will determine the winning and losing manufacturing organizations.