Smart manufacturing systems operate in increasingly dynamic environments, where machines may experience unexpected failures and recoveries during operation. These disruptions pose a fundamental challenge to real-time scheduling, which must proceed without foreknowledge of future jobs or machine outages. In this paper, we introduce a new online scheduling model that incorporates bounded-delay machine failures—unlike classical online scheduling, which assumes static machine availability. Jobs arrive over time and must be assigned immediately to currently available machines. Our objective is to minimize makespan while ensuring resilience to temporary unavailability. We design failure-aware online algorithms and prove competitive bounds relative to an optimal offline scheduler. Our results establish a theoretical foundation for robust, real-time scheduling in failure-prone Industry 4.0 systems.