How can we effectively leverage IoT with data science in a small manufacturing business setting? For a lot of contract manufacturing companies today, the cost of labor is a big factor that influences profitability. The days of off-shoring to China may be coming to a slow natural decline. There are many reasons for this, primarily among them quality (I am sure many of us have experienced a product that was “Made in China” which broke down within a short period of time), cost of transportation, and also political reasons. Additionally, the Chinese themselves are apparently experiencing labor shortage issues. The giant Taiwanese manufacturer Foxconn (which supplies a lot of Apple hardware) had announced that they would install a million robots to improve efficiency and fight labor costs.
This is a natural progression of business, humans are best designed to create and innovate and leave all mundane tasks – such as assembly operations for example, to automation. However, we are not fully there yet. Many small manufacturing companies are highly cash strapped to fully automate their business. They still have to rely on human operators to finish off tasks such as assembly of parts, inspection and so on. For these companies then getting a solid handle on the true cost of labor is critical.
In order to develop actual fully accounted part production costs, they need to literally track how much time a given employee spends at a given workstation or on a specific assembly process. There are several advantages of collecting such data – you can more accurately track your manufacturing overhead costs, for example an hourly worker who is not putting parts together or inspecting or generally involved in the process, is adding to the overhead. Furthermore, if you know that a higher labor cost employee is spending more time working on a lower margin product, you have the opportunity to optimize your production by ensuring that low margin products are handled by lower cost employees. You can extend this to optimizing the overall production scheduling and organization.
Furthermore, you can quickly and easily develop an optimal product pricing strategy by running what-if scenarios using the knowledge accumulated via the data and analytics. If a customer wants you to quote on a new custom product, which requires specific set of operations, you have the statistics to know exactly how much labor effort is going into the product and thus allowing you to know your costs more accurately. The benefits are numerous. But data and the associated analytics are the key to this story.
The first step is to use some form of proximity sensor to detect the presence of a worker near a workstation. The sensor must be calibrated to generate meaningful output which can be used for providing the right analytics. The proximity sensors could be as simple as a combining an arduino board with an Infrared LED or as sophisticated as a GPS-based grid system which covers the entire plant floor area. The key is to make sure that we generate the right kind of output from these sensors so that downstream analytics becomes possible. This is the essence of IoT.
A starting point for the data would be a data table that will tell us, for each workstation, how much time was spent by a given employee, how long was the workstation idle (utilization rate) and so on. When we aggregate such tables from all available workstations in a database, we will have sufficient information for all descriptive, predictive and prescriptive analytics. This is the start of data science with IoT.
One thing this setup assumes is that each workstation is designed to enable only one kind of assembly activity. If that is not the case, then we would need a more extensive internet of things where every part that is worked on has a sensor that feeds information into an on-premise or a cloud database about who worked on the part and for how long. This is what “smart-dust” is supposed to enable. While this may seem somewhat unrealistic and complicated today, the day is not far off when this would become routine. We are literally only a few years away from such tiny sensors! We are already there however in IoT with data science.
Originally posted on Fri, Oct 04, 2013 @ 08:40 AM