- By Paul Darnbrough
- March 31, 2016
- Special Section
- The science of quality management has improved enormously over the past 70+ years.
- Creating a strategy for improvement has to begin by understanding what attributes are critical to customers.
- Some quality elements may be difficult to characterize, and designing a measurement approach can involve a variety of variables.
Understanding what makes a product “good” is key to predictable and reliable production
By Paul Darnbrough, PE, CAP
In nearly all regions and industries, markets for products reflect the impact of globalization. The tactic of competing on price is increasingly giving way to a wider range of factors—including product quality. The idea of improving the quality attributes of products is not new, but it has become far more scientific in its approach. Over the past 75 years, particularly in the time following WWII, people like Joseph M. Juran and W. Edwards Deming developed quality control and management as a discipline.
Quality can be defined many ways, but most definitions focus on customer satisfaction. If a product does what customers need and meets their expectations, it is generally regarded as a quality product. There is no benefit from seeking ways to improve elements outside the customers’ interests.
For a manufacturer, these basic quality requirements must be met at a sustainable price. In fact, to remain competitive, a manufacturer may have to sacrifice some aspect of quality for the sake of the process or underlying economics. All elements ultimately prove to be trade-offs, and appropriate balance must be found on all sides.
The enemy is variability
Every company struggles in one way or another with variability related to products and services. Variability can affect feedstocks and raw materials; it can cause problems with manufacturing processes; and it can ultimately affect how a customer experiences the quality of a given product. It is difficult to imagine how variability would be desirable outside of artistic endeavors. If some variation causes an improvement, a reasonable company looks for ways to make the change permanent. Therefore consistency in all aspects is a worthwhile objective, even if making it happen can be a challenge.
When dealing with quality in the way we now generally understand it, a manufacturer has to know how its customers define quality. What aspects are important to them? Let’s use a basic example: beer. Arguably the most important quality attribute of beer is taste, but there are more subtle aspects, including color, aroma, and carbonation. And to a beer distributor, long shelf life when stored at a reasonably consistent temperature is a critical quality characteristic.
To maintain customer loyalty, most brewers work very hard to make their products consistent. This is not an easy task from year to year when there can be great variability in the agricultural products used as raw materials. Quality managers depend on a combination of laboratory and taste testing at every stage of production. Throughout the supply chain, they work hard to characterize barley malt, hops, adjuncts, and water in an effort to understand how variability changes product characteristics and how they have to compensate to turn out a consistent product in every enormous batch. Throughout the process, highly trained individuals use their senses: sight for color, clarity, and foaming; smell for aroma; and taste for flavor.
Some manufacturing elements, such as temperature, can be controlled by automation, but others depend on human intervention.
What gets measured gets done
The statement above, in various forms, has been attributed to many sources, but the message is simple: any attribute related to quality that is worthy of being monitored or managed needs to be measured, which means manufacturers must select an appropriate measuring device.
If a company makes a relatively simple product and its most critical quality attribute is pH, selecting an appropriate measuring device is straightforward. If the manufacturing process variables affecting pH are understood, data from one or more pH sensors can be tied back to the control element in a closed loop. If the pH value begins to drift from the set point, the control system takes corrective action.
The problem in real-world manufacturing is twofold: key attributes can be difficult to measure, and the related control elements can be difficult to manipulate. Furthermore, it is entirely possible that controlled elements affect multiple attributes in a complex interrelation.
Attributes related to product quality are not normally restricted to the “big four” process measurements: flow, pressure, level, and temperature. Those variables certainly have a significant effect on product quality, but many other measurements related to quality are more analytical in nature and may depend on a mix of attributes. Some of those might be difficult or even impossible to measure using conventional technologies, so manufacturers may have to measure something else as a proxy.
For example, how is it possible to measure the taste of a fruit-juice drink made from a blend of juices, sweetener, and water? The flavor characteristics of fruit change based on source, seasonality, and other circumstances, but the product needs to be consistent. The desire is for a mix of sweet and sour, but taste is subjective and cannot be quantified and measured directly.
However, from an analytical standpoint, suppose the desired flavor is most closely related to the sugar content and acidity of the juices added. Sugar can be measured using density or specific gravity. Acidity can be measured using pH, conductivity, or possibly some other type of analytical sensor.
If the natural content is not within limits, it may need to be corrected in the blending process. These two characteristics become a proxy for taste, and can serve as at least a rough guide. A human taster may have to make a final judgment, but the degree of deviation from an ideal should be reduced.
The problem with a proxy is the possibility it can be influenced by some other aspect of the manufacturing process. If the specific gravity or acidity level can be pushed up or down by something other than the fruit juice characteristics, the quality system might make assumptions based on the reading and try to correct the wrong thing. For example, the acidity level might be correct, but the flavor will not be as desired.
Depending on the nature of the attribute to be measured and its relationship to the proxy, more than one proxy might be required. Getting a more accurate picture of the flavor might require both a conductivity measurement and pH, because while the two are closely related, maintaining both at desired values gives a more nuanced and accurate picture of the product’s flavor.
Real time versus sampling
Some attributes, such as density or conductivity, can be measured continuously in real time. Manufacturers can place a sensor in a pipe or vessel and measure changing values. Other measurements may have to be performed at intervals using samples. Most analyzers—such as tunable diode laser spectrometry, Raman, or near-infrared spectroscopy—work from a sample and may take some time to complete an action. When used in a continuous process, they might indicate the characteristics of the product made 30 minutes ago, so corrections cannot be made to the process in real time. However, this data is still valuable for optimizing the process.
Batch processes are often easier to work with, because they are not constantly changing as with a continuous process. Often batches can be held, and the attributes of a given batch can usually be adjusted to some extent in the manufacturing process. In the earlier example, beer is a batch process, although made in a large facility in huge batches. The brewmasters check the product at strategic points to determine if it is exhibiting the desired characteristics at each step. It is made from variable ingredients, so the recipe may need to be adjusted to compensate for a different character in the malt, water, or other ingredient to reach the ultimate objective of a consistent final product.
Pharmaceutical companies typically have complex definitions of quality, and the mechanisms to measure those characteristics are also highly complex. Most manufacturers are bound by regulatory agencies to monitor their processes and adhere to approved procedures.
Historically, these manufacturers often did not check the quality-related aspects of a product batch until the process was finished, at which point it either passed or failed. In more recent years, manufacturers have been designing programs to support testing at multiple stages throughout a process to ensure all manufacturing elements are coming together as expected to ensure a good final product. Ideally, the number of rejected batches are reduced, because early discovery of a problem allows corrective actions while the batch is still in process.
Matching attribute and sensor
At some point the task may fall to you as an automation engineer to find a way to measure and control some aspect of quality. While many situations are unique, there are some generalities about how to approach such a situation. In process applications, measuring quality often involves some type of chemical composition analysis:
- ensure a component considered a contaminant does not exceed limits
- verify major product components are present in appropriate proportions
- confirm reactions have been completed without excessive levels of unreacted feedstocks
- measure moisture content (or some other liquid) in solids
- ensure separation processes, such as filtration or distillation, have isolated desired products
The first step is determining what the relevant measurements are, and how they are defined and quantified. Some will be very direct and specific. Such clarity might not make them easier to measure, however it will be easy to tell when success has been achieved. Others might be more nuanced, such as the taste and aroma examples cited earlier. Automating something subjective is a major challenge, and part of the design task might be to identify proxies and to design appropriate measurement and control strategies for these parameters.
The ultimate objective, which might not be fully realizable, is to maintain quality attributes with closed-loop control, whereby the automation system manipulates some element of the process to maintain the variable at the desired point. A blending process is a simple example: two or more chemical components are mixed and a strategic characteristic, such as density, is monitored to indicate the proportions are correct. Flow rates of each component can be adjusted as needed following a basic loop strategy. Others might not be so simple.
Some products are highly complex and pose serious challenges for producers. A very common example is gasoline. When a refinery blends feedstocks into this everyday commodity, dozens and even hundreds of individual chemical components might be included. The refiner has to ensure the product meets a variety of specifications, such as octane, sulfur content, and vapor pressure.
With so many components and attributes on the list, finding a simple measurement available in real time that is capable of delivering a definitive answer regarding product quality is elusive. Producers instead rely on a variety of analyzer techniques and peculiar measurement sensors (such as a “knock engine”) to monitor output and verify product characteristics. In the real world, gasoline producers use a combination of real-time and grab-sample analysis, online and in a lab.
Fortunately, most situations are not this complex. Simpler products have simpler quality measurements, although some, such as wine production, thrive on human analysis and resist mechanization. Even the largest-scale beer brewers still depend on humans to ensure uniformity.
The role of data
When data related to quality is difficult to interpret, acquiring more history to draw upon often provides critical insights. When it is possible to look at a long series of batch production records, some connections may emerge that can show cause/effect relationships not immediately apparent in real time. Of course this is typically an offline research activity.
A few basic statistical tools, such as regression analysis, might be able to connect specific manufacturing data with positive or negative quality outcomes. The result could be a new tool to control specific quality attributes, hitherto unrealized. In exceptional cases, a process variable previously regarded as inconsequential might have a significant effect, and controlling it might therefore become critical. As with any statistical analysis, the more good data available, the better, so it is important to use all relevant measurements and other control system information.
Understanding key relationships
Ultimately, manufacturing a quality product depends on a thorough understanding of what quality means from the customer’s viewpoint. It may not be practical to satisfy every possible interpretation of quality, so a manufacturer must weigh trade-offs and make appropriate compromises to satisfy as many interpretations as possible in a sustainable production environment.
Once these desired quality attributes are fully established, the process of linking specific process variables to these attributes can begin. This can take time because those links may not always be obvious. It can take even more time to determine exactly how to measure those variables, since they might not be the typical choices.
But in the right hands, modern automation system components, such as controllers, measurement devices, and final control elements, can usually be successfully applied by engaging good design practices to produce consistent and quality products at reasonable costs.
We want to hear from you! Please send us your comments and questions about this topic to InTechmagazine@isa.org.