Neural Networks Application In Ultrasonic Target Recognition Of Tower Crane Safety Pre-warning

Nowadays, the casualties and economic losses caused by tower crane instability and collision accidents has presented an increasing trend in construction industry, thus, security monitoring and control has become one of the important factors influencing the wide application of the tower crane. Many companies, research institutions and scholars have conducted long-term and in-depth research about monitoring of tower crane instability and anti-collision system. However, all results are still dominated by passively defensive measures, which are lack of flexibility and practicability. Therefore, it becomes extremely urgent to develop smart and active safety pre-warning system of tower crane.Aiming at high technical cost of tower crane monitoring and its lack of intelligentization, initiative, flexibility, speedability as well as real-time performance, the target recognition method based on neural network and ultrasonic signal time sequence was proposed in this paper, which combines advantages of both ultrasonic sensor and neural networks together. By optimizing ultrasonic signal feature extraction, type selection design of neural networks and security early warning system building, the tower crane safety pre-warning of instability and collision prevention was carried out.The focused research contents are as follows:1.The propagation characteristic of ultrasonic sensor, operating characteristic of transducer, relationship between time series of ultrasonic signal and texture, shape and location of the target object were analyzed through experiments. Then, time domain feature of ultrasonic echo and time series were analyzed, so the mapping rule between amplitude of the ultrasonic signal correlation characteristics of time series as well as distance characteristics of time series values and the target geometric and physical characteristics, distance information and the twist angle were obtained, which provided a scientific basis for extraction for ultrasonic signal characteristics.2.According to the above mapping rule, the target recognition method based on neural network time series of ultrasonic signals was proposed. An in-depth study was conducted on the input selection rules of neural network, evaluation methods of network design, sample library structure, learning and generalization selection of design rules, etc. Meanwhile, according to different functional requirements of tower crane safety warning, three kinds of neural network information fusion methods including BP, Elman and SOM were put forward aiming at single and multi ultrasound time series respectively.3.By analyzing stability mechanism, limit load and stability under co-movement of tower crane, the judgment methods of static and kinetic stability were gained. Then, the overturning criteria of early warning and monitoring the instability were obtained through deformation. Additionally, combined with the need for anti-collision construct, the BP, Elman, SOM neural network object recognition methods were applied to different links for the tower crane safety warning, so that the information acquisition, data fusion, proactive early warning function were achieved.In this paper, the target recognition methods based on neural network were applied to ultrasound safety warning systems for tower crane, which presented various improvements in terms of intelligentization, initiative, flexibility, speedability as well as real-time performance compared with traditional methods. For one thing, it can meets the practical requirements of tower crane pre-warning, promoting the safe and wide application of tower crane greatly; for another, the research on methods of target recognition based on neural network and time series fusion can be deployed to other sensors, which can provides references for low-cost sensors to obtain high-accuracy data fusion rapidly.