Outlier anomaly assessment method based on the attention-enhanced LOF algorithm: a case study of loose particle detection signals in aerospace relays

Quality and reliability are essential in aerospace systems. Aerospace relays, as critical components, must pass stringent loose particle detection tests before factory dispatch. This paper introduces the Attention-Enhanced Local Outlier Factor (AE-LOF) algorithm, an advancement of the Local Outlier Factor (LOF) algorithm, specifically designed to improve the detection of anomalies in loose particle detection signals from aerospace relays. This novel algorithm incorporates an attention mechanism to enhance the outlier sensitivity traditionally hindered by the LOF algorithm's dependence on the k-value, which often results in underestimation of anomaly degrees. The AE-LOF algorithm refines the LOF calculation method by integrating outlier evaluation metrics, thus broadening the detection range and capturing more significant outliers. Experimental results demonstrate that AE-LOF exhibits greater sensitivity in detecting outliers, effectively enhancing the detection and recognition of loose particles within aerospace relays. This method proves to be of substantial practical value in the field of aerospace loose particle detection.