The emergence of Industry 4.0 revolution has increased the availability of data from various engineering components providing extensive information on different aspects of the industry. In the context of reliability and efficiency, features extracted from industrial data can be utilized to assess system health status and to optimize product service as well as determining remaining useful life (RUL) through the framework of the prognostics process.</br> This framework is a key feature and therefore connect the operation of systems in higher modes of automation and factory digitalization focusing on the asset life cycle for reliable and continuous operation. Furthermore, this allows to estimate reliability characteristics mainly related to lifetime and reliability's characteristic (hazard rate, availability), thus enables plant operators to manage the maintenance action and related logistics tasks (spare part management) effectively.</br> Hence, the need for prognostics algorithm able to establish correlations between related variables and measured variables becomes obvious. Using the proper algorithm during operation, operational cost, and in the best case due to connected maintenance strategies, unscheduled machine downtimes can be reduced. Through appropriate control strategies, it is possible to preserve the service life based on the information of damage accumulation in unpredicted circumstances.</br> Even with inadequate information extracted from monitoring data, prognostics schemes allow to predict upcoming physical characteristics that permit a higher level of condition-based system maintenance. </br> Commonly, degradation progression is modeled according to the specific configuration using existing algorithms with assumed numbers and state conditions. However, due to the complexity, especially in the case of a system with multiple hidden states, the proper configuration is hard to assign and to identify. Concerning prognostics and health monitoring process, the correlation between measured data and deterioration level must be established for real lifetime estimation. Moreover, for a component with stochastic deterioration, a single model structure might not able to represent the dynamics of degradation. Real-time data are generated during operation, so incomplete data about failure and usage up to the end of life are expected.</br> A model able to describe the nonlinearity of degradation to predict future damage progression for real-time application has to be defined. The need for unsupervised state estimation process to assist degradation modeling preventing under or over assumption is also apparent. Among the existing approaches is the application of clustering methods to classify data and to estimate the number of degradation states might exist. However, integration into the lifetime prediction framework is still infancy and often not considered. Due to that reason, a data-driven prognostics algorithm approach is developed using a state machine concept to realize a model able to deal with the variability and uncertainties of industrial data. </br> This approach provides a systematic solution for a system exposed to multistate degradation by enabling the system to identify appropriate parameters autonomously, according to the state it belongs to. The goal is to identify hidden relationship or correlation between monitoring data with degradation level and design parameters.</br> The selection process is aided using a state machine approach, which will describe the degradation process according to the transition condition. This method is capable of correlating degradation data and consumed lifetime through assessment of system health state and selection of appropriate lifetime model, therefore increasing the flexibility and adaptability to different complexity of measured data. To optimize the configurations, Non-dominated sorting genetic algorithm-II (NSGA-II) is integrated and make full use of prior measured data during the training procedure. The refined state machine model then used for calculation of remaining useful life.</br> Moreover, the establishment of a prognostics approach based on a nonlinear degradation model to enhance prediction performance is also considered. For the accurate prediction of system lifetime, estimation of future degradation from the point of assessment is required. At this point, the unavailable data are numerically calculated by integrating linearized gradients adaptively by considering nonlinearity in current degradation. The coefficients used to dene future degradation gradients are identified according to different states assuming future linear degradation increments.</br> These coefficients are determined using an optimization-based algorithm simultaneously with the calculation of consumed lifetime by extending the previously established state machine lifetime model. To allow optimality in configuring state topology, in this work, a previously developed state machine lifetime model is extended for the optimal number of states identification based on K-means clustering algorithm and cluster validity index.</br> Combining unsupervised state estimation process with a new state machine lifetime model has transformed it into a semi-supervised prognostics approach. </br> Finally, adaptability of the proposed prognostics approach is validated using run-to-failure monitoring data from the tribological experiment, simulation wind turbine degradation, and plant growth agricultural experiment. Due to the flexibility and scalability, this model can easily be deployed to various types of industrial data. According to the findings, from metric accuracy analysis and end of life evaluation, the proposed approach capable of increasing the precision estimation compared to previous models. On top of that, this approach demonstrates the ability to improve structural health assessment and lifetime prediction in a more flexible way to address the variability in the system.</br> Impressive findings obtained highlight the capability of this approach as an alternative strategy for structural health assessment and RUL estimation for various type of industrial application.