COMBINED GENETIC AND SEQUENTIAL QUADRATIC PROGRAMMING ALGORITHMS FOR VIBRATION- BASED STRUCTURAL HEALTH MONITORING

Structural Health Monitoring (SHM) of a structure is a requirement desirable for many
engineering structures in both civil and aerospace applications. Early warning of impeding
disasters, if not identified on time, damage may have serious consequences, both safety
related and economic. However, the complexity of large structures and the difficulty in
accessing them makes the use of traditional Non-Destructive Evaluation (NDE), such as
visual inspection and instrument evaluation methods, impractical. An effective alternative is
the use of methods that depend on Vibration-Based Damage Identification (VBDI). These
methods are based on measurement of changes in the modal characteristic of the structure,
that is, its frequencies and mode shapes, to identify changes in its physical properties.
Therefore, measurement and monitoring of vibration characteristics should theoretically
permit the detection of both the location and severity of damage.
In the present paper, the output error method of damage identification has been used to assess
the presence and severity of damage in steel structures. The identification problem is
formulated as optimization problem to find parameters describing the presence and location
of damaged elements in a structure by minimization of discrepancy of the modal response
predicted by FE analysis from the test data (mixed analytical-experimental approach). A
typical portal frame has been chosen for experimental measurements. The modal parameters
were measured through modal testing. Two approaches are introduced to solve the
identification problem. The first approach deals with a discrete problem that is solved by
Genetic Algorithm (GA). The second treats a mixed discrete-continuous problem that is
solved by a combination of GA and Sequential Quadratic Programming (SQP), in which the
GA is used to produce a high quality starting guess for the (SQP) method that is applied to
accurately identify damage. It is found that the first approach was able to identify damage in
case of moderate and sever simulated damage, while the second approach is capable to
identify damage in case of light damage.