Degradation in lithium-ion battery often shows a macroscopic result of a series of complex physicochemical reactions within the battery, generally manifesting as capacity degradation. In particular, the capacity of NMC batteries degrades in a two-stage during the service cycle, which is linear degradation in the first stage and non-linear rapid degradation in the second stage. Understanding the aging mechanism of non-linear aging is of great significance for predicting battery capacity and controlling battery properly to prolong its service life. The thermodynamic performance degradation mechanisms of lithium-ion batteries are generally classified as loss of lithium inventory (LLI), loss of active material in positive electrode (LAMPE), and loss of active material in negative electrode (LAMNE) in the present research. Currently, the degradation modes diagnosis for an aging battery is to determine the mechanisms that causes the capacity decrease, which includes destructive analysis method and non-destructive analysis method. The destructive analysis method disassembles cells in various aging states to observe and compare the morphology, structure and element changes of these cells. However, the method requires specific equipment and is invasive to the battery, making it unavailable for use in device and only suitable for laboratory research. The non-destructive analysis method extracts the signatures of the battery and obtains the internal information of the battery to achieve the aging diagnosis through the voltage and current data monitored by the battery without disassembling the battery. The above methods can provide insight into battery aging, but the battery cannot be charged with the designed current sequence or removed from the working environment in application. Therefore, an in-operation diagnosis method for degradation modes is proposed in this paper, which only requires partial operation data without special demand on sampling accuracy and frequency. The Extend-Average-Voltage and capacity are proposed as indicators and simulated evolutions with aging under multiple degradation modes. The results reveal that the nonlinear aging is related to the rapid degradation of negative electrode material and lithium-ion inventory when cell resistance increases linearly. The proposed method can achieve 100% diagnosis of non-linear aging modes by experimental data, which is verified by voltage reconstruction model. This work is of significant significance in developing online aging diagnosis model.