Research And Software Development Of Residual Life Prediction Method For Led Lamps

With the continuous development and improvement of semiconductor technology,Light Emitting Diodes(LEDs)have gradually become the current trend in lighting products as the fourth-generation green light source.LEDs have been widely used in various applications such as road traffic,construction,interior decoration,and automotive lighting.The prediction of LED luminaire lifespan aims to forecast the performance degradation trend and remaining service life of system components before failures occur.By analyzing the degradation of illuminance,the current state and future development trend of system components can be determined.Currently,there is limited research on the remaining lifespan of LED luminaires in the domestic market,and long-term lifespan prediction is not possible.To address the issue of poor accuracy in predicting LED remaining lifespan,this paper explores a method for predicting the remaining lifespan of LED luminaires and visualizes the experimental results.The main contents of this paper are as follows:Firstly,statistical analysis methods utilizing the smoothing process algorithm and ARIMA model algorithm were employed to predict the lifespan of LED luminaires.It was found that for short-term predictions,the quadratic smoothing algorithm performed well,achieving the best MAE(Mean Absolute Error)of 94.01.The ARIMA model also exhibited good convergence with an MAE value of 163.69,demonstrating favorable model generalization,compatibility,and tolerance.However,the long-term predictions deviated from the actual situation and yielded poorer results.Furthermore,an analysis of various parameters of LED luminaires was conducted,and a data-driven algorithm model was constructed to achieve long-term residual life prediction for multidimensional time series.A three-channel fused CNN(Convolutional Neural Network)model algorithm was designed to extract features from multidimensional parameters.The extracted features were then input into Bi LSTM(Bidirectional Long Short-Term Memory)for prediction,and an attention mechanism was incorporated to enhance prediction accuracy.The improved CNN-LSTM-Attention model allows the model to efficiently process large amounts of data while focusing on the dynamic performance of the samples.Experimental results revealed that this model algorithm exhibits a high degree of fit for predicting long-term time series data,with an RMSE(Root Mean Square Error)of 14.90 and MAE(Mean Absolute Error)of 7.94.This indicates that the model is suitable for long-term residual life prediction involving large datasets.Lastly,LED luminaire devices were designed,and reasonable aging tests were conducted to analyze the causes of problems and potential failures during the aging process of LED luminaires.The structure of a data acquisition system was designed to achieve real-time data collection and management of LED parameters.Appropriate hardware conditions for equipment control were configured,and data was transmitted to an online health management platform through wireless communication.Python was chosen as the software development tool,utilizing the Py Qt toolkit to design the frontend interface.The backend utilized the Tensor Flow framework to build deep learning algorithm models,enabling the visualization of prediction algorithms.