During the smelting process of the AOD furnace, the unbalanced reaction of material will lead to the occurrence of splashing. It will not only damage the smelting equipment but also seriously injure the personnel. In this study, first, the information of liquid level, audio information and vibration information are detected by multiple sensors respectively. Then the fused information is used to forecast the splashing. Finally, the multitasking fuzzy controller is used to suppress splashing. The results show that the method of forecasting and suppressing splashing can accurately forecast and achieve rapid suppression. Thus the efficiency of smelting can be improved.

The manufacturing process of large-area, high-precision gratings is a very complicated and time-consuming process. The hardness testing of grating films is an important step in the entire process. In order to simplify the manufacturing process of gratings, we have proposed a new method for testing microhardness based on tool edge indentation. Also, it unified tool adjustment and microhardness testing steps in the grating manufacturing process. First, a mathematical model of the relationship between tool load and indentation contour length is established. The model parameters were then modified using tool indentation experiments with different loads. When measured with a nanoindenter, the average hardness of the grating film was 447 MPa. The hardness value of the grating film obtained by our proposed method is almost the same as that measured by the nanoindenter, and the maximum deviation is about 2.2% of the average hardness value. The experimental results show that our proposed method can replace the microhardness test method of using a nanoindenter. Therefore, the disadvantages of using a nanoindenter to test the hardness of a grating film are avoided, such as the limited sample size, the sensitivity of the indenter to the roughness of the film and the depth of the indentation, and the accuracy of film testing, and the efficiency of grating ruling can be improved.

In order to control the end-point carbon content in the process of argon–oxygen refining ferrochromium alloys, according to the mechanism model of the oxygen supply rate and carbon content variation, the mechanism model of the argon supply rate and carbon content variation, and the mechanism model of the supply rate of the two gases and molten iron temperature variation, this paper proposes an improved multivariable inference control system and uses expert control to realize the automatic adjustment of the filter time constant in the multivariable reasoning controller. The simulation results show that this method can ensure the stability of the control system without errors and maintain the furnace temperature below the maximum withstanding temperature, further improving the dynamic response of the system and effectively shortening the smelting time of the entire argon–oxygen refining ferrochromium alloy. It is beneficial for improving smelting efficiency and prolonging furnace service.

As an important factor in mechanical production, the wear state of single crystal diamond tool directly influences the machining quality and dimensional accuracy of the workpiece. In order to improve the recognition accuracy of tool wear state, a multi-information fusion tool wear state recognition method based on Wolf Pack Algorithm (WPA) optimized BP neural network was proposed in this paper. The tool vibration signal and the acoustic emission (AE) signal were used as the tool wear state recognition signal to perform wavelet packet decomposition. In this way, the best characteristic frequency band representing the tool wear state is obtained, and the energy value of the best frequency band obtained by the signal is taken as the characteristic parameter. After normalization, the multi-information WPA-BP tool wear state recognition model is input to realize multi-information fusion and output recognition results. By optimizing the initial weight of BP neural network by WPA algorithm, the problem that BP neural network is prone to fall into local minimum during training is solved. The experimental findings demonstrate that the accuracy of optimized multi-information fusion WPA-BP model recognition reaches 86.7%, which effectively improves the accuracy and stability of tool wear state recognition.