Tuesday, September 20, 2011

Another Thought on Popper: Structural/Mechanical Design

DanQuo's applied science: Another Thought on Popper: Popper’s idea of falsifying rather than verifying a hypothesis is essentially what structural engineers do during the design process. The...

I believe the preceding article describes exactly the process which was criticised by Popper.

The whole structural/mechanical design process described concerns accumulating supporting evidence of suitability. Suitability is determined by comparison with some predefined acceptance criteria: increasingly some code of practice. The proposed structure once assessed against the acceptance criteria and found compliant is then approved for construction. Once constructed it becomes a real world experiment. Once it eventually fails, the acceptance criteria in the codes of practice are revised. {The alternate hypothesis that the beam will fail holds true.}

What I believe Popper was arguing is that need to deliberately go looking for the evidence which will falsify the proposed hypothesis, not wait for it to turn up. The evidence collected in Europe suggests that all swans are white, the evidence in Australia suggests swans are black. What ever the evidence supports it is necessary to go looking for the complement, opposite, alternative or challenging hypothesis. The evidence to date suggests that swans are either black or white: but what evidence is there to support that they cannot be blue, green, yellow or red? Other birds have these colours so what prevents swans from having these colours?

From statistics have the null hypothesis and the alternative hypothesis. The null hypothesis is that the structure is fit-for-function or suitable for purpose. The alternative hypothesis is that the structure is not fit-for-function and it will failure. Quality robust design does not seek to minimize the probability of the failure event but rather deal with the failure.

Traditional structural design was primarily concerned with gravity loads and preventing the structure from sinking into the ground or collapsing under its own self-weight. Real world failures have resulted in increasing requirements for consideration of wind loading, seismic loading and live loading.

The Ronan point disaster, was the consequence of failing to consider a potential failure event. The gas explosion lifted the floor up, and blew the wall out, and the floor then had no wall to sit on. Having a mechanical and manufacturing engineering background, I have always found the structural codes to be some what deficient when it comes to loading considered. I believe there are typically considered to be 6 degrees of freedom, though I have a tool design handbook which identifies 12. There are three axes of translation, and 2 directions along these axes, and three axes of rotation and 2 directions of rotation about.

Structural codes have a tendency to only consider a given axis and only consider movement in one direction relative to. So tradition check gravity loads and prevent movement towards the ground, but otherwise ignore upward wind loading, or ignore upward loading from an explosion. A failure event occurs and codes revised to consider the other direction: thus now consider wind uplift. As I understand recent discussions on SEAint listserver, the international building code (IBC) ignores vertical seismic actions and only considers horizontal, whilst the nuclear industry apparently considers the vertical. However requirements for robustness in structural are starting to introduce more consideration of qualitative acceptance criteria into structural design, if not explicitly identified in the codes. As I understand it good seismic design is more to do with detailing than altogether resisting the magnitude of the forces involved.

Resisting forces is a problem. The alternative hypothesis that the structure is not fit-for-function and will fail holds true: the structure will fail. It is not possible to design earthquake resistant buildings, or hurricane resistant or flood proof buildings. The so called earthquake resistant buildings will be destroyed by earthquake, the hurricane resistant buildings will be destroyed by hurricane, and the flood proof buildings will be damaged by flood. Because structural design is based on supporting evidence rather than falsification these failure events are not considered.

The BP oil rig was always going to fail, and it was always going to leak oil into the Gulf of Mexico. Good design is not to make its failure a low probability event, but to design for the failure event. If failure events are made low probability then people become complacent about the hazards, and emergency services are otherwise cut. What could just be an inconvenience escalates into a disaster. It has been suggested that earthquakes and floods in undeveloped countries are less of a problem than in developed countries. The argument being that the people in the undeveloped countries have less to loose and also consequently less to recover. Whilst the developed countries have highly integrated and interdependent systems of supply: life style and life support is entirely dependent on the infrastructure of the city: loose the infrastructure and life is severely affected. If struggling to survive in the first place, then life after the earthquake, is still the same struggle for survival.

Engineers are sued and/or prosecuted because they put forward false propositions, such as the building is safe or that the building is earthquake resistant, when the building is no such thing, and cannot be made so. The point of moving towards limit state probabilistic design is so as to avoid such false propositions, to be more explicit that the design is based on low probability of failure, not zero probability of failure. What is the probability that a US ship will be boarded by the enemy: until the USS Pueblo was boarded it was zero, afterwards increased to one. The failure event needs to be dealt with, not ignored because of low probability. Good design does not involve increasing the design loads every time a structural failure occurs: it requires consideration of the mode of failure and the response to failure. {NB: not suggesting buildings be designed to resist meteor impacts or aircraft flying into them: the cause of the failure is a different issue.  The issue here is that failure has occurred by what ever cause.}

When the multi story buildings in a crowded city collapse, there is no clear space to escape to, there is little clear space where a person can stand and a collapsing building will not fall on them. Increasing the resistance of the structure doesn't provide protection from the failure level event. There is little merit in strengthening existing hospitals to new higher magnitude loads, if the day after strengthening is complete, an earthquake of greater magnitude occurs and destroys the strengthened buildings. Far better to consider the response for the failure event in the first place. Cities have more buildings not compliant with current codes than are compliant, buildings are typically compliant with older now obsolete codes. Strengthening existing buildings may satisfy insurance companies who do not wish to pay out for replacement and therefore want to minimize such pay out, but it does little to really address safety.

Consider the operation of a submarine without instrumentation, all that the captain knows is that there is a 5% probability that he will operate the submarine at a depth which will collapse the submarine: the magnitude of the collapse depth is unknown, and the operating depth is unknown. The captain is unlikely to operate the submarine. When it comes to buildings people either believe their buildings are resistant to extreme events, or aware that there is potential for failure but not under what circumstances. People are poorly informed relative to the decisions they have to make. It would be preferable that people know that a structure is going to collapse before they start hearing the creaking and feeling the movement. When to shelter in the building and when to evacuate the building or when to evacuate the area altogether? If know when the building or other structure is at risk it is not necessary to strengthen to comply with current codes: simply respond accordingly to the current state of the dynamic environment.

Similarly design structures themselves to respond accordingly to loading events. For example a Mongolian Yurt is unlikely to crush its occuppants if collapsed by an earthquake, whilst a concrete apartment block will. Whilst a yurt unlikely to resist a hurricane, it can be packed up and occupants and dwelling evacuated together.

Additionally to be quality robust the structures need to be fabricated and constructed using processes which have low variability, maintained by processes that have low variability, and otherwise designed to have acceptable performance no matter what the variability of the operating environment. Acceptable performance does not mean equal performance no matter what the operating conditions, but some type and level of performance which is considered acceptable for the operating conditions.

Designing and constructing reinforced concrete apartment block in a region where steel is in short  supply is not quality robust. No matter how much inspection is provided the required steel is not going to get into the structure if it is not available. A welded steel structure is not going to be fabricated by certified welders if none are available. The null hypothesis is that the structure can and will be constructed to the specification, the alternate hypothesis is that it will not. If not built to the specification how defective and hazardous will the structure be when placed into use?

Normal design process is largely built around supporting evidence that the proposal is fit-for-function and can be made to specification. Design is a justification process, not a process of falsifiability or refutability. Interpreting Popper as business as usual for the design process, misses the point, we have to deliberately search for that evidence which refutes the hypothesis that our final design is fit-for-function, and further more design for the failure event. Failure of the production process, failure of the product in service and failure of the maintenance process. Whilst the failure event is not possible to avoid, an appropriate response can be determined: and we return to square one: the null hypothesis that the structure is fit-for-function backed up by our supporting evidence of suitability for our predefined acceptance criteria. At some point have to decide that have done all that is practical with respect to supporting and refuting evidence.

When a decision is made it shouldn't just be on the basis of the benefits obtained but also the detriments incurred. Politicians put forward supporting evidence for the benefits of a proposal but leave out the detriments, the opposing politicians present only the evidence supporting the presence of the detriments. In a court of law the coroner and prosecution will accumulate the supporting evidence that a design was defective, the defence will have to provide evidence supporting fitness-for-function in the face of evidence to the contrary.

As far as I am aware thus far no designer has been held responsible for a design which could not be made fit-for-function using the resources available. Thus far manufacturers and builders are held responsible for not making and supplying to specification: for thus far few have put forth the alternate hypothesis that the specification was not fit-for-manufacture, not fit-for-fabrication and not fit-for-construction.

So to recap: null hypothesis structure or design is fit-for-function. The alternate hypothesis the design is not fit-for-function and will fail. How will it fail, what is an appropriate response? When consider resolved this issue, then have returned back to the null hypothesis so reconsider the alternate hypothesis once again. Repeat whilst practical and refuting ideas available.

The difference between DanQuo's article and what I describe, is subtle. I took DanQuo's description to be the stock standard structural design process, which reaches a predetermined conclusion of fitness-for-function. That is the process is iterated until a design-solution which meets the existing acceptance criteria is met. I am suggesting this process does not involve the falsification process, it is all justification. I am saying that when this conclusion from the routine process has been reached, that is the point at which the alternate hypothesis should be addressed, and should attempt to prove the conclusion is false and the design is not fit-for-function. Further more it can always be proven not fit-for-function, and therefore helpful to be aware of those situations. Failures are important.

TEDxYYC - David Damberger - Learning from Failure - YouTube