In the laboratory, some experiments give encouraging results; other results are discouraging. None of these are ‘trouble’ because according to KiloMentor’s definition, ‘trouble’ is an undesired deviation from what was expected that occurs on scale. Undesirable deviations on-scale cost significant money and there is invariably immediate managerial pressure to quickly assign a cause.
Deviations in quality and quantity are most common. ‘Trouble’ leads to a higher-than-expected cost of goods. That can be fixed by fixing the deviation that caused it or by making other compensating changes rather than fixing the deviation. The former is sometimes the easier course of action. For example, costs can be reduced by removing a bottleneck in a process to reduce equipment and labor costs rather than fixing a deviation in a reaction yield.
When talking about deviations we can distinguish two subtypes.
What I would call a Type 1 deviation is a deviation from a result that has been actually achieved already and that you are trying to reproduce by repeating the protocol exactly, but where the outcome is found to be significantly different. For example, you are repeating a procedure identically, at the same scale, but your outcome is significantly inferior.
What I would call a Type 2 deviation is a deviation from a prior result after making at least one change that one expected and hoped would not affect the result in any deleterious way. It is a deviation from a predicted outcome. For example, using a laboratory protocol you run the procedure in the pilot plant as much the same as possible, predicting a similar quality and yield, but the result is significantly worse.
Often, established processes are subjected to reevaluation and improvement efforts because even a small yield improvement can lead to a significant financial benefit. In an effort to reduce operating costs, less expensive grades of reagents, solvents, or processing aids are employed and the impurities in these reaction components can have an adverse effect on processing leading to deviations in quality that are deviation from the hoped-for result of no change in quality. If these preliminary experiments are done at laboratory scale, an unfavorable result is a disappointment but is not ‘trouble’ but if the change was accidental or unintentional and was made at scale it is a type 2 variant of ‘trouble’.
Type 2 deviations also arise in the initial runs of scale-up. Deviations from a hypothetical result may not be deviations at all. The hypothesis that the cases are sufficiently similar to give the same result may be simply wrong. But if we can do something to restore the wished-for expectations it will be wonderful.
Troubleshooting
Troubleshooting is problem-solving under the gun and at scale.
When Trouble happens and you are called to help, treat it as an emergency; act appropriately. Your value to an organization is likely to be assessed predominantly by your skill at troubleshooting when Trouble comes.
If the trouble stops processing, if possible do not go home until the blockage is removed. Take temporary ownership of the trouble, even if it probably isn’t your fault; assigning blame is not a priority in troubleshooting rather it gets in the way of effective action.
The most common emergency is a failure in an in-process analytical test or a required observation. Processing, following the batch sheet, stops until a decision on how to proceed is made; you may be required to provide input to that decision. In most instances before applying problem-solving methods to a deviation make sure the deviation is real. Check the analytical method and redo the analysis. Can one confirm the deviation using an alternative analytic methodology? Nothing is more frustrating than trying to find the cause of an unexpected deviation that actually does not exist.
The most frequent error in troubleshooting is called ‘jumping to cause’. A hypothesis that might explain the deviation comes to mind and immediately the troubleshooters jump into action to test the hypothesis. Only after the hypothesis is proven false and the deviation is not corrected do the scientists consider other hypotheses. The correct mental process or group protocol is to quickly gather information about the deviating batch and the conforming prior experiments. Write down in a table form what the deviation is and what it is not. Asking WHAT, WHERE, WHEN, HOW and WHEN NOT, WHERE NOT, WHAT NOT and HOW NOT. Then construct as many hypotheses fitting with everything that is being observed as one can. Ask whether the hypotheses fit what is known about the problem. Rank the hypotheses from the most probable cause to the least probable.
When there are many hypotheses, the best strategy initially is to combine several changes that are unlikely to have interactions among themselves so that the deviation would be corrected if any one of the changes is the deviation’s cause. Ordinarily, it is not scientifically preferred to change more than one thing at a time, but here, where a number of the corrective actions can be predicted on the basis of their mode of action to be beneficial or have no effect at all, with no possibility of a negative result, several remediations can be combined. If the unsatisfactory deviation remains after this trial most likely all the hypotheses combined in the test were untrue. The true cause is probably among the remaining untested hypotheses.
Often it may be difficult to assign a most probable cause or give a preference to one hypothesis over another. In this situation, one should prefer to test first the hypotheses that are easiest to fix.
When starting a troubleshooting investigation no-one knows how long it will be before the deviation can be corrected. Often measures are started immediately and in parallel to develop what is called a patch. A patch is an additional step inserted in the process scheme usually to purify off–spec material so that it can be used to carry out further steps of the synthesis. A patch has value even if it does not need to be used (such as when the cause of the deviation is quickly found), because that patch may be used in the future if a different problem arises. The patch constitutes new purification knowledge about the intermediate. No knowledge is wasted. Developing a patch is insurance. Some deviations are not easily amenable to a patch. When the deviation stops the process completely and prevents one from obtaining any product, even a product of reduced purity, there is nothing to purify further. An emulsion or the complete failure of a solid intermediate to crystallize are examples of deviations that halt processing.
The cause of a deviation may have been hidden by a team member to avoid blame. Batch records only report what operators 'say' happened. It is possible that a mistake was recognized by operators almost immediately when it occurred, but it was irreversible. It may have been hidden in preparing the batch record. This makes the troubleshooter’s job harder but it is part of human nature and comes with the territory. Operator errors are most likely if some of the operators are replacements for regular workers. In examining a deviation the questions of who and who not needs to be addressed with reference to the operators who conduct the process. Analytical results are almost always generated automatically and overseen by analysts, not operators. The analytic data are therefore more tamper-proof.
During scale up there should be many more samples taken than after the process is well established. In the early scale-up stage testing should be planned and carried out at as many different points as possible throughout the processing. The samples should be stabilized and saved for later analysis in case there is a deviation. These I call forensic samples. They need not be even looked at when the scale-up experiment goes as expected. At any point in the process step where (i) the mixture is a homogeneous phase and (ii)where the sample will not likely degrade over time under normal storage conditions, consider taking a large enough sample both for analytic testing and also to allow one to continue the processing in the laboratory to see whether the deviation has occurred before or after the sampling point.
When a process step has been finalized and is part of regular production, taking forensic samples can stop. Only retain those that are the in-process checks. It is true that now if one experiences a new deviation one no longer has these forensic samples to help track it down, but one has the advantage of knowing that since successful runs have preceded this new deviation, the deviation is a real difference from prior practice not just a deviation from one’s expectations based on laboratory results.
No comments:
Post a Comment