Imagine an intelligent factory. As a production order comes up, the system checks that the needed parts are in inventory. It confirms, then, the part is not there. Inventory shows it as available; however, the warehouse lacks the current item.
An AI driven system may not help in this case. You need to establish good data practices before trying to implement AI solutions
The problem is indicative of the issues that arise when putting together an intelligent factory.
This article will discuss these difficulties and how to deal with them
1. Why manufacturing data is fragmented
Modern factories often use a mix of technologies and approaches in managing production:
Enterprise Resource Planning (ERP): Tracks business processes, procurement, and inventory management functions at the enterprise level
Manufacturing Execution System (MES): Focuses on production planning and execution, material management, and quality control
SCADA: Supervisory Control and Data Acquisition systems monitor and control industrial equipment
Industrial internet of things (IoT) sensors: Enable operational visibility and control
Radio Frequency Identification (RFID) tags and readers, location tracking devices: Provide information concerning the location and status of tagged equipment, materials, and people
Each of these technologies serves a particular purpose in production management. They often do not interoperate seamlessly within a single framework. This lack of uniformity has consequences. For instance, an ERP system may indicate that a certain part has come into inventory, but it may take hours or days for an MES to recognize it as available for production.
This problem becomes an issue when trying to establish a clear picture of the shop floor status. Each of these technologies is likely providing valuable data, but data quality is difficult to maintain if sources do not share a common framework.
2. Why data quality takes priority over AI
Artificial intelligence applications depend on data collected from operational technologies. When those data are incomplete or wrong, the results are misleading
Take predictive maintenance models, for instance. They require accurate data concerning the timing and nature of changes in equipment condition. These models may fail if timestamps within a data source are not consistent. As such, predictive maintenance does little to help with equipment failures when timestamps are uncertain.
Manufacturers may consider taking one of the following three actions to improve data quality.
Consistent identifiers
Designate a set of identifiers that apply consistently throughout the production lifecycle. Consistency helps align the information from different systems and determines if a particular set of data corresponds to the same item in two different databases.
Reliable data exchange
Standardize the manner in which data are shared between different systems. Specify procedures concerning the frequency with which data updates occur and how the integrity of data exchange is maintained in the event of an interruption. Depending on requirements, systems may communicate via application programming interfaces (API), industrial field bus technologies, events, or middleware applications.
Data validation
Develop procedures that validate the integrity of data. Where appropriate, define rules and procedures for addressing missing data, events that occurred inconsistently, unexpected data values, and other irregularities. In essence, it is critical to ensure that no data make it into the AI models that do not meet specific criteria.
Such actions help ensure that data integrity is consistent and that there is a baseline of quality data with the required degree of completeness and consistency.
3. How to establish real-time operational visibility
Location tracking technologies can help manufacturers obtain an up-to-date view of production operations
Systems such as radio frequency identification (RFID), ultra-wideband (UWB), and others support real-time or near–real-time tracking of containers, tools, equipment, and people. The information helps manufacturers understand the location and status of items and optimize production processes. However, there are limitations to such technologies.
RFID information, for instance, is of limited use unless it can be correlated with production orders. As such, one may consider how this technology can supplement other data to improve overall insight. The combination of RFID and production order information, for instance, informs the materials manager when a particular tag has arrived at the production line and if its contents match a given production order.
The extent to which location technologies contribute depends on the application and environment. Issues such as the need for tagging, interference with other devices, update speed, and tagging power may limit the value of a given technology. As such, manufacturers considering location tracking technologies should address the following questions:
In which context is location tracking applicable?
What level of precision and how frequently does the technology need to update the position of a given item?
How will obstacles and interference affect the performance of a given technology?
What is the technology’s power consumption and lifecycle?
How much does the technology cost to implement and operate?
These considerations help ensure that the chosen technology is appropriate for the intended application.
4. A phased approach to implementing AI solutions
A phased approach may help manufacturers adopt AI applications with limited risks and costs.
The process generally involves the following steps:
Define the problem.
Focus on a specific issue that the proposed solution will address.
Avoid identifying an AI-related solution before determining what the problem is.
Determine the data landscape
Identify what data are needed to address the identified issue and which systems house this data.
At times, this phase reveals roadblocks that necessitate additional preparation before proceeding.
Establish a baseline.
Depending on the identified problem, this step requires establishing a threshold or a point of reference against which improvements will be measured. For instance, if one focuses on reducing the search time involved in finding materials, one would establish the rate at which workers currently search for materials.
This phase is important because it helps establish concrete evidence of improvements rather than assumptions.
Test the solution in a controlled environment
This phase generally involves identifying a small application area or procedure within operations and subjecting it to the proposed solution.
It is important to verify that the data landscape supports the projected results and that the solution delivers on expectations under natural experimental conditions. The target audiences should also test and use the technology under ordinary working conditions.
Scale based on results, costs, and other considerations
Compare the results, implementation costs, and other relevant factors against established expectations. Apply the changes fully if the results are positive and sustainable within the given context. Adopt a wait-and-see attitude if the results are inconclusive.
5. Determining if the solution works
The proposed solutions should ideally contribute towards achieving the initially identified goals.
If an organization focuses on reducing the time it takes to identify materials, for instance, it might measure the average time it takes to manually locate parts or supplies after an order comes in. The organization might also count the proportion of time inventory information is inaccurate.
In either case, the manufacturing organization should measure the extent to which the application has improved productivity in the targeted area. It should further consider other relevant factors, such as the rate of change in production levels to ensure that it measures progress toward overall production goals.
Putting it All Together
Smart manufacturing entails much more than simply adding more technology to the production floor. Even as manufacturers invest in AI applications, they need to identify problems that AI can meaningfully address and ensure that appropriate data exist to support AI applications.
Solving some production-related issues may be as simple as applying good materials management practices. Addressing the issues that affect production requires an understanding of how different data sources relate. For instance, the ability to locate a particular production container can be helpful but insufficient without correlating this information to production orders.
For engineers and manufacturing technology specialists—what has been the biggest challenge you have faced while bringing together these disparate systems and technologies; was it the data quality, technology interoperability, ensuring real-time information exchange, or training and change management on the shop floor?
Top comments (0)