Fatigue Life Assessment of Tower Crane Based on Neural Network to Obtain Stress Spectrum

In order to reduce the probability of crane safety accidents, a method based on radial basis neural network is proposed to quickly obtain the stress spectrum and calculate the remaining life of the crane. Firstly, taking an in-service tower crane as an example, an ANSYS finite element model is established based on actual parameters, and the finite element model is statically analyzed to obtain the location of the dangerous point. Secondly, the typical operating conditions of the crane are simulated. The position of the trolley and the lifting load are used as the input layer while the equivalent stress value at any point is used as the output layer to train the radial basis neural network model. Using the trained radial basis neural network model can obtain time-stress curve at any point quickly. Finally the remaining life is assessed based on the fracture mechanics method. The results show that this method that using the radial basis function neural network model to obtain the time-stress curve at any point can greatly save the cumbersome process and a lot of investment in the field measurement of the crane, and also provides a reliable basis for the long-term safe use and later maintenance of the crane.

Prediction of Surface Roughness of Abrasive Belt Grinding of Superalloy Material Based on RLSOM-RBF

It is difficult to accurately predict the surface roughness of belt grinding with superalloy materials due to the uneven material distribution and complex material processing. In this paper, a radial basis neural network is proposed to predict surface roughness. Firstly, the grinding system of the superalloy belt is introduced. The effects of the material removal process and grinding parameters on the surface roughness in belt grinding were analyzed. Secondly, an RBF neural network is trained by reinforcement learning of a self-organizing mapping method. Finally, the prediction accuracy and simulation results of the proposed method and the traditional prediction method are analyzed using the ten-fold cross method. The results show that the relative error of the improved RLSOM-RBF neural network prediction model is 1.72%, and the R-value of the RLSOM-RBF fitting result is 0.996.

ĐIỀU KHIỂN TRƯỢT THÍCH NGHI SỬ DỤNG MẠNG NƠ-RON RBF HỆ THỐNG BỒN ĐÔI TƯƠNG TÁC

Trong bài báo này, điều khiển thích nghi sử dụng mạng nơ-ron RBF (radial basis neural network) được đề xuất cùng với bài toán giảm chattering trong điều khiển trượt hệ thống bồn đôi tương tác. Mạng nơ-ron RBF được sử dụng để xấp xỉ hàm trong luật điều khiển trượt. Hàm signum trong luật điều khiển trượt được thay thế bởi hàm tanh để kiểm chứng hiệu quả của bài toán giảm chattering. Tính ổn định của giải thuật đề xuất được chứng minh bằng lý thuyết Lyapunov. Để chứng minh hiệu quả của phương pháp đề xuất, các kết quả mô phỏng với MATLAB/Simulink của phương pháp này được so sánh với điều khiển mờ, điều khiển trượt với điều kiện tích phân, điều khiển PID mờ và điều khiển vi tích phân tỷ lệ (PID) truyền thống. Các kết quả so sánh cho thấy rằng, bộ điều khiển đề xuất hiệu quả hơn với thời gian tăng là 0,1271 (s), không có vọt lố, triệt tiêu sai số xác lập, thời gian xác lập là 0,2464 (s) và không xảy ra hiện tượng chattering.

Vibration analysis of an experimental double bridge crane system with artificial neural networks

In crane systems, lifting, carrying and lowering the load from one place have different dynamic effects on the system. One of these dynamic effects is the moving load problem caused by the movement of the load on the crane system. With the increasing technology in recent years, production speeds have increased. For this reason, it has made the requirements for fast-running cranes mandatory for the transportation and loading of products. Therefore, it is important to know the dynamic effects of the moving load in fast working conditions. In this experimental study, the dynamic effects occurring on the crane beams with different loads and different working speeds during the transportation of the load on the crane are analysed. Here, there are multiple cars on the crane, and these cars are designed in different numbers on the crane and can be operated at different speeds. Under these conditions, the dynamic effects that have arisen have been tested. Also, vibration measurements were carried out at different points on the bridges. And then, these parameters obtained were used in two different proposed neural network types to predict the vibrations that occur on the crane system. Simulation results show that two approaches suggested that a radial basis neural network type can be used as an adaptive predictor for such systems in the experimental applications.