In industry, remote condition monitoring systems are usually designed to collect and analyze the operating conditions of critical machines or plants so that their current health states can be revealed before fatal breakdown. For the vast majority of rotating machinery, rolling bearings are one of the most widely used and most likely to fail components. Bearing failure can lead to improper operation of the machine and costly downtime, and ultimately cause the breakdown of the entire machine. Therefore, vibration-based condition monitoring with fault diagnosis is becoming more common in industry to increase machine availability and reliability. Moreover, this type of monitoring allows the machines to be located in remote or hazardous environments and enables the use of moveable machines for which wireless and remote condition monitoring is the only option. Considerable research efforts have recently been directed towards the development of wireless and remote machine condition monitoring. Wireless and remote monitoring often encounters the challenge of needing to transmit large amounts of vibration data collected by sensors. Vibration-based bearing fault diagnosis generally requires a high data sampling rate, therefore, an enormous amount of data is transmitted from the sensors to the remote server (data receiver) so that fault diagnosis can be performed. Wireless transmission of vibration signals faces several bottlenecks, such as limited bandwidth and prolonged power consumption, due to large data transmission. To minimize storage requirements and transmission loads, it is necessary to remove the redundant and irrelevant information often embedded in vibration signals and then transmit the data in a compressed format. Furthermore, the compressed data recovered must have no significant loss of bearing condition data that could reduce the accuracy of fault diagnosis. The raw vibration signal usually has two problems that need to be addressed. First, the signal could be nonlinear and non-stationary. Second, the faulty bearing signal is often overwhelmed by noise and larger vibration signals generated by other machine components, such as the nearby gearbox or a defective shaft. Other popular signal processing methods for analyzing nonlinear and non-stationary signals exist, and the ensemble empirical mode decomposition (EEMD) method has recently become popular. This method is an improved version of the empirical mode decomposition (EMD) method, an adaptive signal processing method for nonlinear and non-stationary signals. EMD automatically decomposes a complex signal into a number of different bands of signal components called intrinsic mode functions (IMFs), each of which theoretically represents a single mode of the decomposed vibration signal. However, many researchers have reported that IMFs may contain more than one mode (i.e. mode mixing), which increases the uncertainty of bearing fault diagnosis. The EEMD method introduces white noise to separate disparate time scales and solve the mode mixing problem of the EMD method. However, the major concern regarding use of the EEMD method is the proper determination of its parameters. In this study, a relative-error-based optimization method was developed to automatically select the appropriate EEMD parameters for the signal to be analyzed. As a result, the special temporal signal component related to the bearing fault can be extracted from the raw vibration signal. Another concern is that the bearing vibration signal is overwhelmed by noise in some real industrial cases. In these cases, the EEMD method does not perform as expected, i.e. the impacts generated by bearing faults cannot be separated from noise and other irrelevant signal components. Hence, a hybrid signal processing method that combines spectral kurtosis (SK) and the EEMD method was designed to recover the bearing signal from large noise. With this method, an optimal band-pass filter based on the SK of the signal is first used to filter the raw vibration signal so that necessary extremes can be recovered for the following signal decomposition. The filtered signal is then decomposed using the optimal EEMD method. After decomposition, the IMF that has the largest correlation coefficient with the raw vibration signal is selected as the resultant signal. The impacts generated by the faulty bearing elements then reside in the selected IMF without interference from large noise, and a relatively clean signal is available for further lossy signal compression and accurate bearing fault diagnosis. In the selected IMF for a faulty bearing, the distribution for most of the extremes is around zero. Almost all meaningful extremes related to the defect are concentrated in a small fraction of the samples. Hence, an IMF-based lossy signal compression method has been designed. The compression method provides a good compression ratio for the transmission of vibration signals. The compressed data fit within a limited bandwidth and enable fast transmission. For wireless and remote bearing condition monitoring, a system was developed to host all of the methods described above and demonstrate the wireless transmission and reception of vibration data from faulty bearings. To verify the effectiveness of these methods and the system, raw vibration signals were collected from various faulty bearings mounted in an experimental motor, a real traction motor, and a machine fault generator. The signal processing and compression methods were applied to the vibration signals captured from those motors. Bearing fault diagnosis was also performed to analyze defect features extracted from the signals. The results of experiments and industrial applications demonstrate that the signal compression system can recover the signal of interest, even when noise levels are high and provide good compression performance for bearing vibration signals. With the help of the system, wireless data communication efficiently transmits multiple channels and large amounts of data. These methods can be used for other industrial data, which often contain high noise levels, and remote machine condition monitoring with limited bandwidth, extending the power of wireless devices by reducing transmission time.