Digital Twin-Based Safety Risk Assessment Of Human-Machine Integrated Manufacturing Cells

Achieving human-machine integration is one of the key technical difficulties in upgrading the manufacturing industry in the context of Made in China2025.In the traditional human-machine manufacturing cell,where robot capabilities are constrained by perception and collaboration technologies,how to improve the robot-human relationship in the framework of digital twin technology to enhance the autonomy and multi-dimensional information interaction capabilities,so as to realize the evolutionary transformation to human-machine co-integrated manufacturing cell is an urgent problem to be solved in the context of China’s manufacturing upgrading.Achieving human-machine integration is one of the key technical difficulties in upgrading the manufacturing industry in the context of Made in China 2025.In the traditional human-machine manufacturing cell,where robot capabilities are constrained by perception and collaboration technologies,how to improve the robothuman relationship in the framework of digital twin technology to enhance the autonomy and multi-dimensional information interaction capabilities,so as to realize the evolutionary transformation to human-machine co-integrated manufacturing cell is an urgent problem to be solved in the context of China’s manufacturing upgrading.Based on the digital twin technology,this paper maps the human-machine manufacturing cell in the physical world to the virtual information space through digital model,and uses the data collected by sensors and cameras in various parts of the robot to predict and control the cell,forming a multi-physical quantity and multi-time scale technology process.It aims to realize intelligent and safe human-machine co-integrated manufacturing cells,and ultimately improve the digital and intelligent production and management of manufacturing enterprises.The difficulty of applying digital twin technology lies in the fact that the data required to build a virtual twin is highly dependent on the upstream and downstream data chains and therefore difficult to form a closed loop,and on the other hand,the heterogeneity of the required data from multiple sources makes it difficult to be processed in a unified manner.In order to solve the above difficulties,this paper constructs separate digital twin models for the robot system and the operator in the human-machine co-integrated manufacturing cell.Under the complex condition that the operator’s pose has high degrees of freedom and uncertainty,accurate human pose modeling is essential to study and analyze the human-machine co-integrated manufacturing cell.Based on this,this paper focuses on algorithms related to human posture recognition and prediction,combining digital twin and human-machine cointegration safety risk assessment framework,and the main work is as follows.(1)The Double-Blaze Pose human spatial pose recognition algorithm is designed.(1)The Double-Blaze Pose human posture recognition algorithm is designed,combining the deep learning-based Blaze Pose human posture recognition algorithm with the binocular computer vision algorithm,taking the images captured by the binocular camera as input,using two sets of Blaze Pose algorithms to independently process the binocular images to obtain the coordinates of two sets of human posture key points,and finally correcting the spatial depth coordinates of human posture key points with binocular parallax to achieve accurate Finally,the spatial depth coordinates of human posture key points are corrected by binocular parallax to achieve accurate human posture estimation.In the human-machine integration manufacturing scenario built in this paper,the designed algorithm greatly improves the accuracy of estimating the depth coordinates of human key points compared with other methods,and alleviates the drawback that the previously designed algorithm is susceptible to pose and environment interference,which makes it difficult to accurately infer the spatial depth information of pose.(2)The Pose-Fusion human posture prediction model is designed.In order to achieve "prior control" of safety risks in operator-robot interaction,accurate prediction of the operator’s posture movements will effectively avoid potential safety accidents.The human posture prediction algorithm designed in this paper uses 1D-CNN to extract the spatial features of the data,combines Long Short-term Memory(LSTM)with Attention to form a temporal network layer to extract the kinetic features implied by the temporal changes of the data,and finally,the fully connected Finally,the fully connected network outputs the prediction results based on the extracted highdimensional features.Thanks to the Double-Blaze Pose human pose recognition algorithm,the Pose-Fusion human pose recognition prediction model does not need to recognize videos or pictures,but only needs to use the spatial coordinates of pose key points output by the Double-Blaze Pose human pose recognition algorithm as the input of the prediction algorithm,which ensures the real-time prediction process.In the research related to human-machine co-integration,the United States and Japan,as manufacturing powerhouses,occupy the leading position,while China starts late and lags behind the international one by about 15 years in terms of time span.This paper proposes a digital twin-based safety risk assessment method for human-machine co-integrated manufacturing units,and tests and verifies its effectiveness in a laboratory environment,with a view to exploring the vast blue ocean of humanmachine co-integration in the manufacturing industry in the context of Made in China2025.