Name: SAP AI ; Predictive Maintenance ; Series Overview
Uploaded: 2023-9-7
Duration: 612 s

Share feedback

<p><u><strong>Planned maintenance of Practice Systems</strong></u></p>

Planned maintenance of Practice Systems

Why create an SAP account?

Enhance your learning experience

SAP Learning

Choose your subscription

Manage your subscription

Live Sessions

Hands-on Practice

Certification

My Learning

Ready to start learning?

<div><div><div>In this playlist by Tahir Hussain Babar (Bob), from the SAP HANA Academy, we use SAP Artificial Intelligence to build a Deep Learning model on Acoustical Sound for Predictive Maintenance with SAP AI Core. In this video, we go through an Overview, and look at the highlights of the video series. <br><br>00:00 - Intro<br>00:27 - Playlist Overview<br>02:36 - Digital Audio Sound<br>03:29 - Acoustic Features<br>04:58 - Machine Sound Classification<br>06:38 - Playlist Video Highlights<br><br><br>SAP AI in BTP: <a href="https://www.sap.com/products/artificial-intelligence.html" rel="nofollow noopener noreferrer" target="_blank">https://www.sap.com/products/artificial-intelligence.html</a><br><br>Please note:<br>                     This video was originally published on the SAP HANA Academy YouTube Channel                      and therefore may contain references to SAP HANA Academy and YouTube.                      The tech world is always evolving and the SAP HANA Academy YouTube channel                      is set to close on December 31, 2023.                      At that time, references will no longer be valid                      and will gradually be removed when the videos are updated.<br>Also, this video includes automatically generated captions                      and therefore might misrepresent the spoken content or display incorrect product names.</div></div></div>

SAP AI ; Predictive Maintenance ; Series Overview

Welcome to the SAP HANA Academy. My name is Bob, and in this video series, we'll be looking at SAP Artificial Intelligence. In this video playlist we'll be building a deep learning model on acoustical sound for predictive maintenance with SAP AI Core. In this first video of the series we'll be going through an overview of the whole playlist. I can imagine what you're thinking with the title of this video series called Predictive Maintenance. Well again as explained in the overview what we'll be covering is Predictive Maintenance with Sound. So what does that mean? Well imagine we have a factory and in that factory we manufacture certain products. Now, if there's an issue with one of our production machines, of course we want to identify that issue as quickly as possible. Because if there's an issue with the machine, it might continuously churn out damaged products, which of course we want to stop. So in the playlist scenario we'll imagine that we have a factory and one of the machines has a microphone that continuously records sound. We'll know the sound we're expecting from that machine, but if there's an issue with the manufacturing process, maybe there'll be a sound that you don't recognize, so it's an abnormal sound. Examples of those abnormal sounds could be as follows. You can imagine we'll have an expert who works in a factory that knows what the normal sound of the machine should be. And they can tell that with that normal sound everything is working fine. However, if there's an abnormal sound it could mean that there are some frictions between various parts. Of course there could be some issues with the machine and it could be that some maintenance is required. So we'll imagine that this expert has built kind of a categorization of various sounds. We'll have normal sounds and we'll have other sounds which are not normal or abnormal. And in our example, the expert has classified the different sounds as OK, those sounds that have anomalies and another different type of anomaly in a photo called anomaly 2. So how are we going to do this? So let's first establish what a digital audio signal looks like. So as you probably know, sound is a pressure wave that propagates through air. So here we can see a waveform. On the x-axis we've got time and on the y-axis we've got the amplitude. Now when we place a microphone to record sound, you're basically sampling the amplitude of the wave at discrete time intervals. So you'll end up with a series of numbers which represent the amplitude of the pressure wave. Sound files also come with a few fundamental bits of metadata, such as the sampling rate, bits per sample and the number of channels. So for example, a CD is recorded at 41.1 kHz at 16-bit and it's in stereo. In terms of acoustic features of sound, it can be represented in two different domains. One called the time domain will measure the amplitude, and the time domain example is a waveform. The other example is called a frequency domain and this is where we measure frequency over time. And spectrograms are a good example of frequency domains. Again, instead of measuring amplitude, they measure frequency. Now for both of these types of representation, machine learning experts can extract numerical features that can be used as inputs to do some analysis. Whether you do machine learning or perform deep learning. Now in deep learning, as well as utilizing these spectrograms, there are also other variations of spectrograms, such as log-mel spectrograms and something called MFCCs, which stands for Mel Frequency Cepstural Coefficients. So these are basically bi-dimensional maps, so they can actually be seen as images. And since we reduce sounds to an image, one of the techniques that we can use to analyze our sounds, of course, which have been interpreted as images, is those convolutional neural networks that we also used in the automated defects detection playlist. So how are we going to classify the sounds that come from that machine? Well, we're going to build a model using TensorFlow. So again, we're going to imagine that our expert who works on the machine floor has categorized the sounds that come from the machines. So we'll have different kinds of sounds. Basically, we'll have expected sounds which you could consider normal and we'll have two other categories of abnormal sounds. And because we've got this expert classifying the different types of sounds, it's considered a supervised approach or it's considered acoustic scene classification. Now there's other types of unsupervised approach, but we'll be focusing on that acoustic scene classification. Now, all of these audio clips are exactly 1.5 seconds long and we're going to utilize a Python open library called Labrosa to load our audio clips. Again, we're going to use Labrosa to read the acoustic features of those WAV files. So we'll have an image, or it can be considered an image, which is an interpretation of each of those sounds. So here you can see that the OK sounds have a different look to the sounds which have anomalies. Afterwards, what we're going to do is build our convolutional neural network model using TensorFlow. And lastly, we're going to look at training and testing that model with a small data set. So how are we going to build this model? Well, that's what we're going to cover in the rest of this playlist. So I'm going to take the next two or three minutes to highlight the main parts of this video series. In the next video I'm actually going to give you a whole overview of how we develop the model, but we're going to do all of the work just in a Jupyter Notebook. And this is how your data scientist would develop a model in the real world. After we've looked at how the model was developed, then what we'll do is we'll look at getting our data source or those sound files into an S3 bucket. Then we'll be introduced to the SAP AI Launchpad focusing on the SAP AI Core administration tab. We'll need to do some setup here. After that we're then going to look at orchestration. So we're going to look at building Docker containers in order to process the training and the serving of our model. Now within these Docker containers exists your code in order to train and serve your model. So we'll also have a look at the Python code, and when we've understood that Python code, we'll then look at building and also pushing these containers to our Docker repository. Now in order for SAP AI Core to be able to access and use these Docker containers, there's a couple of configuration files in YAML format we'll need to look at. And after we've done this, we'll create a configuration that can access our data source, which are those sound files in the S3 bucket. Now comes the part of executing the training of our model. So we'll have a video on how we can utilize those configurations. And after your model has been trained, you're going to want to test the accuracy and the performance of those models in case you want to improve them. So we'll look at some metrics. Now, assuming you're happy with the accuracy of your model, then what we'll look at is deploying that model, so you can utilize the model to make predictions. And what are those predictions? Well, we're going to be using Mailman and a Jupyter Notebook. And what we'll do is we'll select a random sound wave and SAP AI Core will indicate if the sound wave is OK or normal or if it has an abnormal sound. And what we'll do at the end of the video series is we'll select three random sound files and we'll determine which ones are normal and which ones have those abnormalities. We'll be doing this also in mailman which basically means your apps will be able to connect and do the same thing. Whether those apps are built in SAP Business Technology Platform or elsewhere. So if you like what you see you should move onwards to the next video in the playlist where we look at how the model was developed. We hope you enjoy the video series and thanks for watching.