IBM Predictive Quality

Predictive Quality is a feature of Predictive Solutions Foundation on Cloud that does the following things:

  • Detects unfavorable changes in the quality of components (inspection entity)
  • Detects unfavorable changes in variable-type data, and provides information that facilitates diagnostics and alarm prioritization (parametric entity)

IBM® Research's Quality Early Warning System (QEWS) algorithm in Predictive Quality detects emerging quality problems sooner and with fewer false alarms than is typically achieved by traditional statistical process control. To achieve earlier detection, QEWS is sensitive to subtle changes in data values, such as shifts that are small in magnitude or trends that grow slowly over time. For a given level of statistical confidence, QEWS typically needs fewer data points than traditional statistical process control.

Early detection of quality problems is essential where delayed detection can have significant negative consequences, such as in the following scenarios:
  • Building a large inventory of defective products results in high scrap costs.
  • Having widespread quality or reliability problems in the field results in damage to brand value.
  • Compromised production of supply-constrained materials or components prevents on-time shipment.
  • Compromised production of products with long manufacturing times results in shipment delays.

Inspection entity

In a manufacturing environment, defects can occur in a manufacturing process because of variations in factors like process, raw materials, design, and technology. The resulting low quality of products creates a larger inventory of defective lots, which leads to increased inspection effort.

A small delay in detecting a quality problem can result in large costs, lost opportunity, and lost brand value.

In the inspection entity, QEWS evaluates evidence to determine whether the rate of failures is at an acceptable level. QEWS highlights combinations for which the evidence exceeds a specified threshold. QEWS can detect emerging trends earlier than traditional statistical process control, such as trend analysis. QEWS maintains a specified low rate of false alarms. Post-warning analysis of charts and tables identifies the point of origin, the nature and severity of the problem, and the current state of the process.

The inspection entity analyzes data from the inspection, testing, or measurement of a product or process operation over time. The data can be obtained from the following sources:
  • Suppliers (for example, the final manufacturing test yield of a procured assembly)
  • Manufacturing operations (for example, the acceptance rate for a dimensional check of a machined component)
  • Customers (for example, survey satisfaction ratings)

Products are the subjects of QEWS analyses. A product is typically a part or a part assembly, but it can also be a process or a material. Products might be used in larger finished assemblies, which QEWS calls resources. A product can be associated with any resource, process, material, location, or a combination of these entities during inspection analysis.

You can adjust the frequency at which data is captured and input to QEWS, and the frequency at which QEWS analyses are run, according to the requirements of each situation. For example, monitoring the quality levels of assemblies that are procured from a supplier might best be done on a weekly basis; monitoring the quality levels of units that are moving through a manufacturing operation might best be done on daily basis.

Parametric entity

In the parametric entity, Quality Early Warning System for Variable Data (QEWSV) monitors variable-type data. Variables are defined for every operation per tool. Variables are equated with Measurement type, whose measurements are read at different time intervals during the sequence of the operation flow. This type of data is found in several industrial applications, including Supply Chain, Manufacturing, and Finance applications.

QEWSV identifies unfavorable trends in the data process. The focus is on providing timely detection of unacceptable process behavior while maintaining a pre-specified low rate of false alarms.

Variable values and evidence charts are plotted by using parametric results. The deviation or drift from target values are computed and analyzed to show whether the process sequence is adhering to normal operation limits.

Predictive Quality handles various master data sets, from end products to manufacturing machinery to the raw materials used, as well as environment or location-specific data. Predictive Quality identifies the following sub use cases. Sub use cases can apply to a combination of different masters or to a lone master entity.

Process resource validation
This category is the default use case, where the process and the resource that takes part in the process is monitored based on a defined set of variables. These variables are associated with a set of parameters that define the target values, acceptable limit, unacceptable limit, standard deviation, false alarm rate, and unacceptable factor.
Resource validation
A resource is monitored based on the standard operation limits across a few measurement types (variables). This type of health check is essential in identifying any issues in the resource and correcting those issues to improve the performance and throughput.
Product validation
With Quality inspection, the product as a whole is checked, based on the failure rate. In variable data, given the set of variables whose targets are set for the product to meet, any deviation or drift beyond the allowed deviation highlights a flaw in the product.
Material validation
Raw materials that are purchased from a vendor are monitored for a defined set of guidelines as variables, and validated to check on the quality of the procured material.
Location suitability
With variable analysis, a location is analyzed to see whether it is suitable for a particular operation. Variables like pressure, temperature, humidity, and their time slide values can forecast the suitability of a location for carrying out any operation.