مطالعه ی آزمایشگاهی تشخیص آسیب در پل خرپایی تحت بار متحرک با استفاده از شبکه ی عصبی مصنوعی و تبدیل موجک تجربی

عوامل بسیاری می‌تواند در طول زمان بهره‌برداری از سازه‌ها منجر به ایجاد آسیب‌های موضعی شود، که در پاسخ‌های ارتعاشی سازه‌ها نمود پیدا می‌کنند. با هدف تشخیص وجود و تعیین موقعیت مکانی آسیب با استفاده از شبکه‌ی عصبی مصنوعی و تبدیل موجک تجربی تحت بار متحرک، یک پل خرپایی در مقیاس آزمایشگاهی ساخته شد. پس از اندازه‌گیری پاسخ‌های ارتعاشی پل، 17 ویژگی حوزه‌ی زمانی استخراج شده از سیگنال‌های خام، برای تشخیص وجود آسیب به‌عنوان ورودی به شبکه‌ی عصبی ارائه و وضعیت سلامت پل (سالم یا آسیب دیده) به‌عنوان تارگت‌های آن در نظر گرفته شدند. برای مکانیابی آسیب نیز 5 ویژگی غیرپارامتریک،شامل: آنتروپی‌های شانون و تی‌سالیس، جذر مجموع مربعات، عامل شکل و کشیدگی، به همراه ویژگی انرژی، از سیگنال‌های پردازش شده‌ی پل به کمک تبدیل موجک تجربی استخراج و استفاده شدند. نتایج نشان می‌دهند که روش پیشنهادی به‌طور موثری قادر به تشخیص وجود و مکان آسیب در پل تحت بار متحرک است.

experimental study on damage detection of a truss bridge under moving load using artificial neural network and empirical wavelet transform

civil structures are always considered one of the most valuable properties of each country. many factors can lead to local damages in different parts of structures during their operational life. these damages are reflected in the vibration responses of structures. this research aims to detect the existence and determine the location of damage in a truss bridge under a moving load using an artificial neural network and experimental wavelet transform. for this purpose, a two-dimensional truss bridge was built in the laboratory to investigate this research’s objectives. earlier experimental studies in damage detection were subjected to excitations such as impact loads and electrodynamic shakers. since the appearance of damage effects in the vibration responses of the structure mainly depends on the applied location of the impact load, a moving load that crosses the entire length of the bridge can be used as input excitation to detect the presence and location of damages for which there is no available data. after measuring the vibration responses of the bridge, 17 time-domain features were extracted from the raw signals, which were used to detect the presence of damage. although feature extraction is applied to raw signals, the signal processing stage was not eliminated for damage localization. by processing the response signals of the healthy and damaged state of the bridge using experimental wavelet transform, these signals were decomposed into different modes and 5 non-parametric damage-sensitive features such as shannon and tsallis entropies, root mean square (rms), shape factor and kurtosis which are all based on statistical parameters in addition to energy, were extracted. finally, these damage-sensitive features were presented as input to the neural network whereas the state of the bridge (healthy or damaged) was considered as its target. the obtained results showed that the proposed method can effectively detect the presence and the location of the damage in the truss bridge.