Abstract: Multi-crack initiation and coalescence phenomena in friction stir welding zones of rocket fuel tanks constitute a critical threat to structural integrity, demanding precise analytical approaches. To address this challenge, the present investigation develops a novel methodology for multi-crack propagation and coalescence analysis in tank structures, integrating finite element simulation with hybrid uncertainty quantification theory while incorporating multi-source uncertainties. The proposed framework employs commercial finite element analysis software to construct a high-fidelity tank model, simulating the complete propagation and coalescence process of collinear multi-cracks within bottom weld zones. This simulation systematically elucidates dynamic crack evolution behaviour, quantifies stress intensity factor variation patterns, and reveals underlying mechanisms through which crack-sensitive parameters govern weld fatigue life. Given the significant impact of multi-source uncertainties on propagation life predictions, a sophisticated hybrid random-interval analysis model is formulated wherein material property uncertainties are characterized probabilistically as random variables, while dimensional uncertainties associated with multi-crack size parameters are represented non-probabilistically as interval variables. This dual-characterization approach rigorously captures hybrid uncertainty effects through computationally derived interval boundaries of structural response and their stochastic features. To overcome computational bottlenecks in uncertainty quantification, an artificial intelligence-based solution is implemented: a BP neural network surrogate model is trained to replace resource-intensive finite element simulations, enabling rapid yet accurate prediction of collinear multi-crack propagation life under multi-source uncertainties. Comprehensive validation via an engineering case study focusing on collinear multi-crack coalescence and propagation in actual welded tank structures conclusively demonstrates the method's efficacy, establishing it as a reliable and computationally efficient paradigm for analysing complex multi-crack behaviour in aerospace fuel tanks operating under uncertain conditions.