Automatic inspection of surface defects is crucial in industries for real-time applications. Nowadays, computer vision-based approaches have been successfully employed. However, most of the existing works need a large number of training samples to achieve satisfactory classification results, while collecting massive training datasets is labor-intensive and financially costly. Moreover, most of them obtain high accuracy at the expense of high latency, and are thus not suitable for real-time applications. In this work, a novel Concurrent Convolutional Neural Network (ConCNN) with different image scales is proposed, which is light-weighted and easy to deploy for real-time defect classification applications. To evaluate the performance of ConCNN, the NEU-CLS dataset is used in our experiments. Simulation results demonstrate that ConCNN performs better than other state-of-the-art approaches considering accuracy and latency for steel surface defect classification. Specifically, ConCNN achieves as high as 98.89% classification accuracy with only around $5.58$ ms latency over low training cost.

Infeed rotary swaging is a cold forming production technique to reduce the diameter of axisymmetric components. The forming is achieved discontinuously by a series of radial strokes that are spread over the shell of the part. Due to tolerances within the rotary swaging machine, these strokes perform individually and the resulting stroke pattern is not homogeneous with regards to circumferential and longitudinal distribution. Nevertheless, in combination with the high number of performed strokes and the large contact area between the dies and the part, the external part properties, such as diameter, roundness and surface roughness, show even values along the finished part. In contrast, strength-defining internal part properties, like microstructure and residual stress components, are more sensitive to the actual pattern and temporal sequence of the individual strokes, which is investigated in this study. The impact of process fluctuations during the conventional process, which are induced by the tolerances of machine tool components, was verified by numerical simulations, physical tests and measurements of residual stress distributions at the surface and at depth. Furthermore, a method is introduced to maintain the stroke following angle ∆φ at zero by flat dies, and thus the actual pattern and temporal sequence of the strokes was homogenized. The results show that the residual stress fluctuations at the surface could be controlled and reduced. Furthermore, it is demonstrated that the depth profile of the residual stresses at a distance of 300 µm from the surface developed independently from the process fluctuations.

In metal forming processes, the forging load as well as the mechanical and geometrical properties of the final product are affected by the actual material flow. Especially for incremental open die forming processes like rotary swaging, the material flow is very complex. By a profound knowledge about how to control this material flow during the process, the final product properties can be controlled. By using FEM, the material flow during infeed rotary swaging of tubes without mandrel was investigated by the history of the plastic strain components (PE_i) in different directions. The influence of the friction coefficient value on the material flow for steel tubes was studied. By the history of plastic strain components tracked during the process, the material flow was analyzed in different radial positions. Based on that, the influence of friction conditions on PE_i was discussed during one single stroke. It was found that with an increasing friction coefficient value, the radial plastic strain firstly increased and then decreased, while the axial plastic strain behaved just the other way round. Thus, the material flow is significantly affected by friction conditions. The simulation results were validated by measurements of wall thickness values of physically formed tubes.

Rotary swaging is an incremental metal forming process widely used to reduce the cross–section of parts. For tubular parts, the final wall thickness also changes during the process. The lubricant condition is a factor, which affects these geometry changes. Beneath the change of the geometry, the complex material flow during the process determines the final geometry and the mechanical properties. Therefore, with a thorough insight into the material flow, it could be understood how to control it in order to achieve desired properties. Producing tubes with uniform outer diameter and changing inner profiles is an application of this method. Furthermore, applying this method, different local cold hardening could be achieved by different total strain. In this study, the dependency of the material flow on the lubrication conditions was investigated. Simulations with combined hardening material models were verified by the change of the wall thickness of tubes. It was found that friction condition significantly influences the back shifting of the workpiece and the elongation caused by each stroke. Results from simulations and experiments showed that a certain lubricant condition leads to the highest axial elongation of the workpiece.

In bulk forming processes, the net shape of a final product is achieved by plastic deformation as the material flows from the initial shape to the final shape of the workpiece. The material flow during the process is an important issue for its relationship with forging force, heat generation, microstructure transformation and energy consumption. Hence, the final properties of the product are directly influenced. Former researches showed that the material flow in the rotary swaging process is affected by different processing parameters like die angle, feeding velocity and friction condition. Thus, a profound knowledge of detailed material flow during the process is essential for a better understanding of the process. By using FEM, the material flow was investigated by the history of the plastic strain (PEEQ) development. In this study a 2D-axisymmetric model was built by using ABAQUS explicit. Both aluminum alloy (3.3206) and steel (1.0308) are studied with different feeding velocities and coefficients of friction. To achieve the development of PEEQ in different areas, the workpiece was divided into radial layers. The PEEQ history of each layer was tracked during the quasi-static forming process. Based on that, the plastic strain rate (PSR) was calculated and examined in a single stroke of the process. In that way, the material flow in different layers is presented and the material flow on the surface differs from that in the center, just the first 1/4 radial area from the surface is sensitive to different friction conditions.