13.2.2 Lessons Learned
The final report of the Committee of Inquiry into the collapse of the Nicoll Highway was submitted on May 11，2005.Faced with the monumental task of piecing together information from 193 witnesses and volumes of documents，the COI elucidated the crucial errors and persons responsible for the collapse of Nicoll Highway.In the COI's words：“Warnings of the approaching collapse were present from an early stage but these were not taken seriously”.It concluded that“the death of four persons was the direct result of the collapse”and that“the Nicoll Highway collapse could have been prevented”.The following recommendations were made by the Committee of Inquiry to avoid similar incidents in the future.
（1）Effective risk management
The potential for major accidents，whether due to the construction process or deficiencies in design，must be recognized and expeditiously controlled.It is inappropriate to leave the control of risk wholly to contractors.Owner's and builder's management must seek a balance between production pressures and quality and safety.
（2）Robustness of design
A robust design is essential.This robustness is provided by identifying the hazards and checking that the proposed design can adequately withstand them.The design should have sufficient redundancy to prevent a catastrophic collapse in the event of a failure of any particular element.Temporary works for deep excavation should be given the same respect as permanent works.
（3）Numerical modeling in geotechnical design
Numerical analysis should supplement and not supplant sound engineering judgment and practice.Those who perform geotechnical numerical analysis must have a fundamental knowledge of soil mechanics and a clear understanding of numerical modeling.
（4）Back analysis
A proper back analysis should be done with an understanding of why the design is not performing as originally predicted，and not just to increase the design levels.The input parameters（e.g.，soil properties）should be adjusted to allow for the displacements and forces to fit the measurements.If these adjusted input parameters fall outside the meaningful range，the analysis model is probably wrong，and should not be used for further predictions.