Machine Learning — Diagnosing faults in the vehicle

Introduction

Many logistics and transportation companies use modules to track their vehicles, and it is normal for some of these modules to not work properly. For instance, the tracker could have to be turned off, erroneous information from the vehicle might be collected, or remote technical help would have to be sent to determine the problem. Because of this, it is essential to run a remote fault diagnostic on certain modules because a failure might result in losses.

The goal of this research is to leverage the data that these modules continuously supply to a database to find module defects. It was necessary to create a system for pre-processing the data acquired and transmitted by the vehicle modules. Following that, models created using machine learning methods were compared and evaluated. However, what is the issue?

Become a Machine Learning expert with a single program. Go through 360DigiTMG's Machine Learning and AI Courses in Bangalore Enroll today!

A fault is an unallowable deviation from at least one of the system's properties or distinctive parameters. Additionally, there are three steps in the fault diagnosis process:

• Fault Detection: Fault Detection is the most fundamental step in fault diagnosis, utilized to look for system failure or malfunction and to establish when the fault occurred;
• Fault Isolation: The purpose of the Fault isolation is to identify the fault's location or the damaged component;
• Fault Identification: The kind, format, and magnitude of the problem are all determined using the identifier.

What is Predictive Maintenance?

Predictive maintenance is performed using data-driven techniques to predict equipment failure and minimise unplanned downtime. It uses a variety of data from in-vehicle sensors, previous service records, and machine learning algorithms to examine various equipment states that might lead to a likely system or equipment failure. The user and the automaker/maintainer are then informed by the ML system that a certain system or component needs to be repaired or replaced.

We can utilise a part for longer because to this strategy's increased component lifespan and decreased unscheduled repair costs.

Additionally, ML-powered quality control systems may be built up to check for any faults in components prior to installation.

Machine Learning Course is a promising career option. Enroll in Top Institutes for Machine Learning in Chennai offered by 360DigiTMG to become a successful.

Why is Predictive Maintenance important?

Numerous advantages of predictive maintenance include stock management, lifetime optimization, recycling management, and optimization of spare part production. For fleet operators, particularly transportation and logistics firms for which downtime is prohibitively expensive, predictive maintenance is another important objective.

Why is Predictive Maintenance important?

Predictive maintenance is a pretty difficult process since it includes data collecting from a range of the system's sensors in order to provide a real-time perspective of the health and reliability of industrial machinery. This maintenance schedule is developed in four steps.

Collecting data from different sensors of the system.

Data preprocessing 

Faults diagnosis and prognosis.

Decision-making on the maintenance strategy.

The study of fault identification and prognosis is of interest to both the academic and corporate communities. A defect is located, isolated, and pinpointed using diagnostics.

• Feature extraction and selection: in this phase, the discriminating features of the raw data are extracted and selected
• Classification of faults: the main task of this phase is to classify the different faults and identify the causes of the failure using the selected discriminating characteristics.

Want to learn more about Machine Learning Course. Enroll in this Machine Learning Training Institute in Hyderabad to do so.

The goal of prognostics is to monitor alterations in a system's operating characteristics throughout the course of its usual operational cycle. You may use it to figure out your equipment's RUL and predict faults before they occur.

Typically, it is carried out in three major steps:

• Construction of the health indicator (HI): the HIs are indexes constructed to represent the health of the equipment.
• Health stage (HS) division: the life of the equipment is divided into different HS based on the defined HI index.
• Prediction of the machinery RUL: the RUL can be estimated through the evaluation of the health status of the equipment

The process I: Feature Set Preparation

The technique looks at two distinct kinds of features: the discrete events that happen during a typical machine cycle and the values of continuous variables then. The raw data is a time series of each digital and analogue IO retrieved from the PLC. First, distinct occurrences are categorised and listed. The discreteness or analogies of the time series variables, as well as the initial event in the sequential machine cycle, must be explicitly determined. Then, one new feature is added to each continuous variable for each discrete occurrence in a cycle. Its output is a feature set that combines discrete occurrences with continuous variables.

• Discrete events that occur in a machine cycle;
• The value of the continuous variables during an interval from the beginning of the cycle;
• And the generated feature set.

Process II: Dataset Preparation

The feature set created by the aforementioned process to generate the dataset; I am merged with the information obtained from the PLC data of the machine.

• Each n discrete event delay feature is filled with the time elapsed between its occurrence and the initial cycle event;
• For each n discrete event delay feature, k continuous variables features are filled with their current value when occurs.
• All missing data are filled with a negative number with -1 since all valid values are positive.

Results

The outcomes of using the ML technique are shown and discussed in this section. For each simulated system, experiments were conducted using two datasets: the PA dataset, which was created according to Process II of the suggested technique, and the ODE dataset, which was created using just discrete timed events. Investigations confirmed:

• The overall performance of the selected models and the influence of combining discrete and continuous variables
• The performance by class of the selected models using a confusion matrix
• The relevance of timed-events and continuous variables features from a feature importance analysis
• The impact of the dataset size on the performance of the models using the F1 Score

Conclusions

A new ML technique for real-time defect detection and diagnosis (RT-FDD) in discrete manufacturing machines (DMMs) is presented and verified using two case studies, a simulated furnace and a pick-and-place machine. Trees and Random Forest are the classifiers selected for the Furnace and Pick and Place Machines.

The mean F1 Score significantly improves by 6% using the recommended strategy for merging continuous and discrete variables in this study compared to a dataset made up entirely of timed-delay discrete occurrences. The statistical difference between the distributions is confirmed by the use of a hypothesis test with a 95% confidence range. The feature significance analysis also reveals that six continuous variables are in the top ten (10) most important features, confirming the positive effects of the dataset's combination of continuous variables with discrete events.

When just 20 examples of each fault class are given, the classifiers utilised with the PA dataset in both case studies show an F1 Score of more than 80%. The F1 Score rises to above 90% with 70 samples of each defective condition. These findings show that the method may identify problems both at initial deployment and when it is still in use.

Future study should consider the inclusion of resources for categorising and grouping special situations, as well as automated anomaly detection for identifying fresh deficient circumstances. Research gaps and new avenues for overcoming barriers to the general adoption of ML on the shop floor may potentially be revealed by new studies that take into account the deployment of AutoMLs in industry and their impacts.