Abstract: An interval prediction method combining a particle swarm optimization（PSO）-optimized Elman neural network with quantile regression was proposed to achieve precise wear prediction for diesel engine piston ring−cylinder liner pairs. Wear data under different operating conditions were first collected using a reciprocating accelerated wear test rig, and a predictive model was constructed with load, temperature, and duration as inputs and wear amount as the output. Key hyperparameters of the Elman network, such as the number of hidden layer neurons, learning rate, and regularization coefficient, were then automatically optimized using the PSO algorithm, while a Dropout mechanism was introduced to enhance model generalization. Quantile regression was integrated to construct a 95% confidence interval for uncertainty quantification. Experimental results showed that, compared to the baseline Elman and Bayesian optimization algorithm Elman（BOA-Elman） models, the PSO-Elman model achieved superior prediction accuracy on test sets for both the piston ring and cylinder liner. Coefficients of determination （R²） reached 0.985 4 and 0.987 9, and mean absolute percentage errors（MAPE） were reduced to 1.14% and 1.12%, respectively. Furthermore, the proposed model attained a prediction interval coverage probability of up to 96.15%, with more compact interval widths, effectively balancing prediction accuracy and uncertainty while ensuring reliability. The method can provide support for the life assessment and condition-based maintenance of piston ring−cylinder liner assemblies.