Quality engineering for a liquid chromatography system - HP 1050 Series Liquid Chromatography System - technical

Hewlett-Packard Journal,  April, 1990  by Helge Schrenker,  Wolfgang Wilde

• E-mail
• Print
• Link

[Figures have been omitted]

For the HP 1050 Series LC system, customer expectations were translated into measurable quality goals, which were then verified by special test methods.

CUSTOMER SATISFACTION is the ultimate benchmark for product quality. This sounds straightforward, but it is far from a trivial task to translate the customer's voice into product quality. It requires finding out, in quantitative and measurable terms, what quality properties the majority of potential customers want, translating these customer expectations into terms, measures, and goals that are meaningful to design and manufacturing engineers, and assuring throughout the design and transfer phases that these goals will be met by the future product.

For two product generations, we have been using the same set of quality criteria. With each generation, we have refined the measures and test methods. Fig. 1 shows the key quality criteria and measures that were applied to the HP 1050 Series Liquid Chromatography System. Using some of these criteria and measures as examples, we will briefly describe how specific quality goals for the HP 1050 Series were set and verified through specially developed test methods.

Setting Quality Goals

Design goals for each of the five quality criteria listed in Fig. 1 were set using three information sources:

* A specific, well-aimed customer survey

* Inputs from senior sales and service people on their perceptions of customer expectations

* Analysis of presently available LC instrumentation

(strengths, weaknesses, potential for improvements).

The customer survey was aimed at gathering current customer expectation data on such sensitive criteria as reliability, uptime, measurement reproducibility, cost for service and maintenance, and so on. Since our resources were limited, we decided to survey only a selected, representative sample of about 100 individuals from our customer base. Our experiences with such small samples in earlier surveys were good. The advantages are low cost and the possibility of enhancing the return rate and results by a mix of mail, telephone, and personal survey methods.

We achieved excellent consistency of survey results. The responses, even to sensitive questions, were very consistent, and compared well with the perception of senior HP field and marketing people and with the results of a thorough technical analysis of a current similar product. (The technical analysis was an analysis of the failure modes of a current product assuming a complete redesign with elimination of all known failure mechanisms for which solutions seemed feasible.)

It was interesting to learn how dramatically customer expectations for the useful life of the product can change from generation to generation. Expectations for product lifetime were more than three times as long as the equivalent data from a survey we had conducted about nine years ago. Setting the right goal for useful life is critical for mechanical assemblies. Higher-lifetime components are often more expensive, so overdesign must be avoided. On the other hand, early wear-out means dissatisfied customers.

One important aspect of surveys is the statistical evaluation of the response data. In this survey, the responses of customers from different markets and different environments for such criteria as expected reliability, service cost, and the like were consistent enough to justify calculating the mean and confidence interval. This was not so for some other results, for example the expected analysis precision, where the responses differed widely based on the respondents' major applications. This was not surprising, since a QC analysis of a drug may require a precision level of 0.3%, while in an analysis of a biological sample in a complex matrix, which requires several sample preparation steps that are all prone to statistical variation, overall expected analysis precision may be only 5 to 10%. In such a situation it is meaningful to plot a cumulative distribution of the survey results as shown in Fig. 2. This plot indicates what percentage of potential users could accept a product designed for a certain performance level (assuming that users with lower performance requirements would accept the product as long as the price were acceptable). This plot is useful for cost/performance optimization.

Similarly, design goals were defined for each of the quality criteria listed in Fig. 1, resulting in a set of benchmarks for HP 1050 Series quality.

Reliability Verification

There are two main approaches to strife (stress + life) testing. The first is the test-to-fail philosophy: the stress applied to a product is increased until a failure occurs. The aim of this test type is to find design weaknesses using high stress levels. The second approach is a strife test with fixed stress conditions (constant acceleration factor*) at a more moderate stress level, which most likely finds only failure mechanisms potentially occurring under normal operating conditions. The focus is not only on finding design weaknesses but also on being able to predict reliability (annualized failure rate, or AFR). On the product and system levels, we do mainly the second type of strife testing.