Prediction of bridge Life-cycle cost of inspections and classification using neural net work

Like human lives , bridges requires specific type of inspection and cares over a certain period of time to preserve and maintain safety for the public usage. however, not all type of bridges require the same type inspection in terms of the time sequence , frequency of inspection , and cost over its life cycle.

Explanatory data analysis

• 6001700 different type of bridges were collected with 137 different features from 1800 to 2018 in which about 16% of data were missing.
• The differential demand in cost of concrete and steel material reveals how the inspection cost can be dependent on the material used for the bridge construction

• The distribution of the kind of material across the US bridges shows how concrete remains the most common and used material in bridge construction. however the cost of inspection isn’t solely limited to the kind of material in use

Material Use Distribution

• The daily traffic frequency significantly impacts the life cycle inspection cost for different kind of material . The steel continuous gives a sense of how the this kind of material is applied for gigantic traffic related bridge projects.

Daily Traffic Distribution across different kind of bridge

• The maximal allowable load and actual load capacity ratio defines the degree to which a bridge is more likely to be subject to collapse when the actual load is being applied . The more the ratio gets closer to 1 the more the bridge becomes unsafe for the public use.

The maximal allowable load and actual load capacity ratio

Modeling

• Predicting the life cycle cost of inspection for a specific kind of bridge using neural network requires putting into consideration continuous variables as inputs and a continuous output variable. the amount of hidden layers depends on the complexity of the problem and the degree of correlation existing among the variables to be captured by the model.
• Rectified linear united (ReLu) is defined as an activation function which transforms input variables and outputs a predictive continuous value.

• The model performance is evaluated by how the training and the testing curves are close to each other and the mean square error is an evaluation metric applied to quantify errors for continuous outcome.

The bridge life cycle cost inspection is expected to be 5 millions $

Multi-classification of bridges on the basis of inspection cost

• In case of multi- classification, the activation function applied to resolve this kind of problem is a SoftMax function which outputs multiple outcomes ( multiple classes) depending on the probability distribution of each class attached to weight .

The model evaluation applies the categorical cross entropy as a metric to quantify classification errors.

Based on results , the life cycle cost of inspection of steel is expected to be 28% higher than the life cycle cost inspection of concrete. although choosing concrete as a material of construction can be costly compared to steel for a specific project , it’s way more cost-efficient when considering the life cycle aspect of each material.