Research On Anomaly Detection Of Friction Stir Welding Process Based On Density Clustering Algorithm

With the growing development of the structural industry chain under the positioning of "Industry 4.0" and "Manufacturing 2025",it is an important goal to achieve intelligent manufacturing in each traditional manufacturing industry,and welding is a very important area of manufacturing.Friction stir welding has unique advantages compared to fusion welding and has been widely used in vehicles,aerospace and other fields.With the development of big data technology and artificial intelligence,if the appropriate use of computer-related technology in stir friction welding,making stir friction welding technology more intelligent,is a major development in the achievement of intelligent manufacturing.In such a background,this article designs a multi-sensor coupled stir friction welding process parameter acquisition system,investigates a machine learning algorithm with improved density clustering,establishes a model based on the data,and uses the model to build a stir friction welding process parameter anomaly detection system.In this paper,a hardware platform for the acquisition of stir friction welding process parameters is designed.Suitable sensing devices were selected for the collection of tool tilt angle,plunge depth and welding temperature.The welding machine at the workshop site is surveyed,and the connection between the sensor and the welding machine is designed to integrate the acquisition platform equipment so that the system does not prevent the stirring head from working while collecting the stirring friction welding process parameters.The hardware integration architecture in the whole acquisition system was designed by using the IPC as the upper supervisor and the sensors as the lower machines,and interconnecting the host and slaves through the matching of communication protocols.The design of the software for the acquisition of stir friction welding process parameters is completed.Based on the completed hardware integration architecture as well as the RS232 and RS485 communication protocols,the software design requirements were developed and the serial communication software was successfully developed in C++.The software includes a series of functions such as the initialization of the serial port,the interaction between the computer and sensors,signal processing,etc.It can determine whether to perform parameter acquisition by judging the idle or working state of the machine,the spindle-driven stirring head rotation in,out,dwell and normal feed state,without manual operation.A channel is established between the serial communication software and the relational database management system My SQL,enabling localized storage of the processed data.An in-depth study was conducted for density-based clustering algorithms.The shortcomings of the DBSCAN clustering algorithm are analyzed,and the PCA dimensionality reduction is targeted to improve the efficiency of the algorithm modeling by using kernel density estimation to adaptively select two important input parameters:domain radius and density threshold,reducing the manual tuning steps in the subsequent modeling process.The performance of the improved algorithm was tested using some artificial data sets,and it was verified that the performance of the improved algorithm was improved.In addition,based on the LOF local anomaly factor algorithm,an anomaly detection mechanism for the welding process parameters was established,and the larger the LOF value obtained by the tested sample through the model,the greater the anomaly likelihood,transforming the anomaly detection dichotomy problem into an estimate of the degree of anomaly.Based on the improved DBSCAN algorithm and LOF,a model and anomaly detection system on the friction stir welding process parameters is established.The system was applied to the welding production workshop and verified that the model can effectively encompass the welding process parameters of the same workpiece.The ability of the system to detect anomalies in the friction stir welding process was verified through the collection of data sets where flying edges affect normal collection.Finally,the reliability of the system design in this paper was verified by the system's detection of the weld temperature variation caused by the undulation of the amount of downward pressure on the shoulder that occurs in actual working conditions.