The journey to implementing artificial intelligence and machine learning solutions requires solving a lot of common challenges that routinely crop up in digital systems: updating legacy systems, eliminating batch processes and using innovative technologies that are grounded in AI/ML to improve the customer experience in ways that seemed like science fiction just a few years ago.

It is a physically larger diameter magnifier, and the result while clear, shows that the magnifier is not intended for detailed PCB inspection. Also, there is an incandescent bulb fitted in the handle (the type of bulb that was popular in flashlights – it has a glass molded lens as part of the bulb), but it doesn’t illuminate the correct area very well. The photo below shows the PCB being illuminated from the rear, just to get a clearer view.

In the meeting room, it’s just a few of us: my manager and two other engineers from my new team, and one engineer from the Recommendations team. We start off with some introductions, and then move on to discuss an issue from the week prior. Evidently, there was some kind of overnight batch failure last week, and they’re still feeling the effects of that.

But perhaps that graph model isn’t too far off from what we need to do here? In fact, the approach described above is a machine learning (ML) technique known as “collaborative filtering.” Essentially, collaborative filtering is an approach that examines the similarity of certain data objects based on activity with other users, and it enables us to make predictions based on that data. In our case, we will be implicitly collecting cart/order data from our customer base, and we will use it to make better product recommendations to increase online sales.

But we’d have to have some way to “rank” the products, because the mapping of one product to every other purchased at the same time by any of our 200 million customers is going to get big, fast. So we can rank them by the number of times they appear in an order. A graph of this system might look something like what is shown below in Figure 1.

DataStax provides the real-time vector data tools that generative AI apps need, with seamless integration with developers' stacks of choice.

For the lens I went with a recommendation on a forum and selected a Computar zoom lens model MLH-10X. The zoom capability is extremely useful. Fixed focal length alternatives include Fujinon (Fujifilm) lenses – they have very nice fixed focal length lenses intended for machine vision. An adapter to go from CS- to C-mount is also needed, because the Point Grey camera has a CS-mount and many machine vision lenses have a C-mount.

To illustrate this evolution, let’s follow a hypothetical contractor who was hired to help implement AI/ML solutions at a big-box retailer. This is the first in a series of articles that will detail important aspects of the journey to AI/ML.

In use it has to be placed almost against the PCB, and then your eye can be anything from around 10cm to right up against the loupe, and the target will remain at a good level of magnification. The solder blob is clearly visible. Also, notice how the PCB on the right of the photo is blurred, yet on the left side of the photo all the stuff on the table is still in focus. This is because the Mitutoyo loupe is less straining on the eyes. Plus the level of magnification is the highest out of all the ones tested. It is a really great choice.

The Meiji EMT has a 4x objective setting too if even greater magnification is needed. Microscopes can be connected to a camera with an appropriate adapter. If a digital single lens reflex (DSLR) camera is available then the lens mount is standards-based so there is a strong possibility that an adapter could be purchased. Otherwise it is sometimes possible to build your own adapter.

For instant results the camera can be directly connected to a PC with a GigEth interface and the camera is powered up with an external power supply.

As for the latter, our Pulsar consumer will write to a Cassandra table (shown in Figure 2) designed simply to hold entries for each product in the order. The product then has a row for all of the other products from that and other orders:

The binocular microscope is good, but I still wanted something more convenient to use directly from the work bench. There are fancy microscope stands, but I didn’t want to spend that kind of money on an experiment. Besides, I felt the binocular microscope was overkill for home use.

We can then take the top four results and put them into the product recommendations table, ready for the recommendation service to query by ‘product_id’:

Looking at the next magnifier from the left, the photo below is the view through the super-cheap e-bay special (just a few $). It has two LEDs on the underside, but they have poor color rendering as you can see from the photo, which makes it quite hard to get a clear view of things. I would suggest that this is second-worst magnifier in the lineup (the worst is yet to come).

If you want the property permanently set in Linux (i.e. to survive after a reboot) then type the line above into the /etc/rc.local file just before the ‘exit 0’ line at the end of the file.

It seems like the current product recommendations are driven by data collected from customer orders. With each order, there’s a new association between the products ordered, which is recorded. When customers view product pages, they can get recommendations based on how many other customers bought the current product alongside different products.

And about things being dirty as ths level, I know right? It's really scary to see things that dirty amidst the COVID-19 breakout.

Fortunately, we did come up with a way to augment the existing, sluggish batch process using Pulsar to feed the cart-add events to be processed in real time. Once we get a feel for how this system performs in the long run, we should consider shutting down the legacy batch process. The PE acknowledged that we made good progress with the new solution, and, better yet, that we have also begun to lay the groundwork to eliminate some technical debt. In the end, everyone feels good about that.

Attempts to upgrade the graphics driver did not result in an improved frame rate for my particular application, and video tearing effects occurred unless the display resolution was reduced. I suspect this is due to limited RAM but I will continue to investigate. Another reason is that perhaps the application is still using old libraries that are in the library path. It is for further investigation but for now I’m happy with the 10Hz frame rate – it allows me to proceed with SMD work easily. I will revisit it and consult with AMD if possible because I'm sure a higher frame rate is achievable (Timesys managed it in their supplied demo for instance).

The other problem, I begin to explain, is that the dataset driving each recommendation is almost always a couple of days old. While this might not be a big deal in the grand scheme of things, if the recommendation data could be made to be more up to date, it would benefit the short-term promotions that marketing runs.

The camera uses a Sony IMX136LQJ sensor (PDF overview) which is quite large (1/2.8” format, 6.4mm diagonal so it will work well with lenses intended for 1/3” and 1/2" format sensors) and has up to 12-bit resolution for very good quality video. It is of reasonable price and it has full HD resolution at 27fps (ideal for video recording) and a Gigabit Ethernet interface for raw uncompressed video capability. With uncompressed HD quality at a good frame rate (more than 25fps) more than 80Mbytes/sec of data needs to be transferred so USB 2.0 is out the window, and Gigabit Ethernet becomes a highly attractive option.

With the flycap software running, the stats at the bottom of the window should say that the Proc/Disp/Req values are approximately 27Hz / 5Hz / 27Hz. This signifies good network throughput, but that the graphics driver is causing low frame rate. I was able to achieve 10Hz by reducing the requested frame rate in the configuration settings in the Point Grey flycap software. 10Hz is not bad for microscope use.

My reason to go for a zoom lens is that it is highly desirable to go from a zoomed-out view on a PCB and then quickly drill-down to a smaller area by adjusting the zoom. Also, with machine vision lenses it is possible to quickly lock the aperture and focus settings with a thumbscrew, and a benefit of this was found to be that you can adjust the zoom by feeling for the lens and only one ring will turn (the zoom ring) without taking your eyes off the PCB.

The next step was to integrate the camera into the home network. A Gigabit Ethernet switch is needed with jumbo frame capability (not all support this, so definitely check! Also, don’t assume that the capability is enabled on the switch by default – usually it needs to be configured).

Magnifiers and other visual inspection tools are essential with working with small electronics. This blog post does a quick round-up of several low-cost ways to achieve this for hobbyists, and shows how to assemble a digital video microscope using the Gizmo 2 single board computerGizmo 2 single board computer (SBC).  See video below!:

Ethernet Controllers Ethernet ICs are available at Mouser Electronics. Mouser offers inventory, pricing, & datasheets for Ethernet Controllers Ethernet ICs.

We’ll want to be sure to examine how our data is being written to our Cassandra tables in the long term. This way we can get ahead of potential problems related to things like unbound row growth and in-place updates.

This post was slightly longer than expected, but the key points are that some handheld magnifiers are actually quite usable for occasional SMD work. Construction of an uncompressed HD microscope system for PCB inspection was presented, using mainly off-the-shelf components and the Gizmo 2 running Gigabit Ethernet.

Some type of stand is required; with a zoomed-in picture the camera would be unusable without one. The camera has M2 and M3 screw holes so a piece of plastic was drilled appropriately (or a 3D printer could be used). The photo below shows the bits and pieces used to construct the camera head.

After modeling this out and running it on our graph database with real volumes of data, I quickly realized that this isn’t going to work. The traversal from one product to nearby customers to their products and computing the products that appear most takes somewhere in the neighborhood of 10 seconds. Essentially, we’ve “punted” on the two-day batch problem, to have each lookup putting the traversal latency precisely where we don’t want it: in front of the customer.

By default the Gigabit Ethernet interface will not perform very well. After jumbo frames have been enabled (see earlier) some performance tuning is also needed and this is achieved by appending the following lines to the /etc/sysctl.conf file:

When working with protoboard and stripboard often the work can be done with the naked eye. This is not always the case when working with modern tiny surface mount components. Basically, they are very small!

How the results are collected and served, though, is a different story altogether. Essentially, the results of associations between products (purchased together) are compiled during a MapReduce job. This is the batch process which failed last week. While this batch process has never been fast, it has become slower and more brittle over time. In fact, sometimes the process takes two or even three days to run.

I often make do with a small handheld magnifier with a built-in lamp and it works ok for the larger SMD devices. I have tried many different types of handheld magnifiers in the past.

The Point Grey camera has Linux software called flycap which can be installed (the steps are described at their website) and the installation on Gizmo 2 was straightforward.

Definition. Diffused light sources refer to lighting that softens shadows and reduces harsh contrasts, creating a more even illumination across a surface.

Although the Point Grey camera comes with application software called Flycap, it is good to get Aravis functioning too because it allows for custom software modifications if desired. The Aravis install steps described here were followed and it ran successfully.

by KCHJ Smolders · 2014 · Cited by 358 — In this study, we investigate whether a higher illuminance level particularly benefits individuals who suffer from mental fatigue.

Cool idea with the ipad! : ) I thought of something similar-ish too, but have not got around to implementing it yet, but maybe it will give some ideas: I was planning to use a tablet PC, since Surface Pro 4 tablets are low-cost used. I picked one up from ebay for dedicating to this task.That way, all software is on the tablet, reducing latency, i.e. no remote viewer software. It does not have a Ethernet socket however, so a USB 3 to Ethernet adapter is needed (these are low-cost).

Some additional heuristic filters may be necessary to add as well, like a “do not recommend” list. This is because there are some products that our customers will buy either once or infrequently, and recommending them will only take space away from other products that they are much more likely to buy on impulse. For example, recommending a purchase of something from our appliance division such as a washing machine is not likely to yield an “impulse buy.”

DataStax serves production-level AI applications with a vector database fueled by real-time, real-life customer and streaming data.

Although not intended for integrated circuit die examination, this is the view looking inside a Motorola 68HC11 microcontroller (a point-and-shoot camera with macro capability would provide a better static photo, but this photo is from a video stream of course:

The image quality was good. The Computar lens is positioned 15cm or much further away from the PCB so it won’t interfere with large components on the board or other items on the work bench. Good to strong lighting is still needed for a noise-free image but the camera is still usable with general home lighting and no task lighting switched on. With the aperture wide open the depth of field is still actually not bad – improves as the aperture is reduced at the expense of more noise of course.

Philips is a world-leading lighting brand that provides sustainable commercial lighting solutions. Efficient, high quality LED luminaires, lamps, ...

I installed Ubuntu (release 14.04.1, not 14.04.2 – the latter has a bug which impacts the installation of development packages so it is better to first install 14.04.1 and then upgrade).

It sounds like that batch process has been in place for about 10 years. The engineer who designed it has moved on, not many people in the organization really understand it, and nobody wants to touch it.

Here is an example PCB with what looks like a fault, as seen through this magnifier. I’ve tried to take the photograph to reflect what I see through the magnifier with the naked eye. So, I’ve positioned the camera very close (about 5cm away) from the magnifier, and it has been manually focussed to a relatively short distance which is what I would do to try and get more magnification out of it. The red circle shows the fault – something like a blob of solder bridging two pins of a 0.5mm pitch integrated circuit. It can be clearly seen, although with a bit of eye strain. I think this is a good magnifier, provided your eyesight is not bad.

My Mum always put picked Blackberries into a jug of water before eating to flush out the 'little critters'. I must admit to not bothering and just eating this 'wild' fruit. I'm still alive.

Hehe everyone is learning how to wash now it seems : ) A friend was telling me people keep coming into his store asking if he sells hand-wash. He keeps stating what should be obvious to them "er no this is a pet store!".

Usually the magnification is quite low, so one is forced to use it close-up to the eyes in order to get a bit more magnification. Possibly this is not healthy! My favorite is the one on the left in the photo (incidentally amongst the cheapest, under $10). It has an LED on the underside and it is angled just right to be very useful. The batteries last a very long time – I’ve not had to change them for a couple of years.

Another future improvement would be to implement a real-time AI/ML platform like Kaskada to handle both the product relationship streaming and to serve the recommendation data to the service directly.

A binocular microscope is another option. This one was a very old model Meiji EMT that I got for next to nothing and cleaned up:

I’ve also experimented in the past with so-called “dental inspection cameras” available from ebay at very low cost (so not the same quality as the cameras used by real dentists I hope!). They are electric-toothbrush-sized and have a sideways pointing camera surrounded by white LEDs:

Universal clip-on LED ring light Compatible with phones, tablets, laptops, and more Padded clip fits screens up to 0.6" thick Improves quality of photos and ...

We’ll start out by putting in an event streaming solution like Apache Pulsar. That way, all new cart activity is put on a Pulsar topic, where it is consumed and sent to both the underlying batch database as well as to help train our real-time ML model.

If your point-and-shoot camera has ‘macro’ capability then chances are that it can be useful to some degree for PCB inspection. A small tripod and good lighting can be useful, and the aperture turned down slightly to gain some depth of field. I got reasonable results with a point-and-shoot camera. The only problem is that most compact cameras that have some HDMI capability will not display a live view via HDMI; the HDMI capability is usually only for viewing photos and recordings. Therefore a photo can be taken of the PCB and then inspected either on the built-in display or from an attached PC or via HDMI as a static image or recording.

First of all, let’s look at data collection. Adding an extra service call to the shopping “place order” function isn’t too big of a deal. In fact, it already exists; it’s just that data gets stored in a database and processed later. Make no mistake: We still want to include the batch processing. But we’ll also want to process that cart data in real time, so we can feed it right back into the online data set and use it immediately afterward.

I didn’t want to have to dedicate a PC for microscope use, so it was decided to try to use the Gizmo 2 for this task. The Gizmo 2 is a small (compact disk sized footprint) single board computer (SBC) with some advantages that made it suitable for the task. The Gizmo 2 has a Gigabit Ethernet interface which is needed but it also has a 64-bit x86 processor and this is needed when running GigE cameras that are supplied with x86 drivers. It is also possible to connect up plenty of local storage which is useful when storing HD video. Check out the Gizmo 2 getting started blog post which explains how to install a solid state drive (SSD) and install Linux.

The pins can be separated by 0.4mm or less, component markings are just as small, and flux deposits make it hard to view any detail until the board is immaculate.

These cameras come with C-mount or CS-mount fittings which are standards that define the screw thead and diameter and the distance from the mount to the sensor so that lenses can be interchanged between cameras. It was easy to find an off-the-shelf adapter to connect between the C-mount camera and the microscope from ebay. The particular camera in the photo above is standard definition (and used and monochrome – I don’t think it would be worth spending money on a color standard definition camera nowadays – see below for HD cameras) and has an analog (composite) output that can be directly connected to a monitor so that I don’t have to use a PC for it. Otherwise it is possible to connect to a PC for recording images or video by using a video capture card.

Sometimes I'll try to dust off PCBs with an air spray can if I need a very close-up photo, but that's not recommended due to risk of static, so isn't a general-purpose solution.

Note that Microsoft Windows can auto-assign the network interface an IP address and so will the camera; thus no DHCP nor static IP addressing is needed; if you test with Mac or Linux, do share the procedure in the comments below.

The camera stand assembly is shown below (the camera is inspecting its own Gizmo 2! – the blue and yellow cables go to a switch – see text further below). The assembly was mainly constructed from 0.5” diameter steel posts from Edmund Optics. Best practice is to use 1” or thicker for such designs, but 0.5” is much cheaper and worked fine. It is extremely stable, and can be rapidly swung out of the way if not needed.

Jumbo frames are packets of data (payload and header, also known as an Ethernet Frame) with a larger size than that defined in the Ethernet standard. Throughput can be increased by using jumbo frames because less header content is transferred on average compared to the payload volume.

So, after further research I obtained a Point Grey camera from ClearView Imaging. I selected this camera because of the good sensor size and sensitivity. It is physically quite small and cube-shaped – about 3cm on each edge:

Premier Farnell Ltd, registered in England and Wales (no 00876412), registered office: Farnell House, Forge Lane, Leeds LS12 2NE.

You don't have permission to edit metadata of this video. Edit media Dimensions x Small Medium Large Custom Subject (required) Brief Description Tags (separated by comma) Video visibility in search results Visible Hidden Parent content Building a Digital PCB Inspection Microscope with the Gizmo 2 Poster Upload Preview Ok Cancel

For now I'm using a head-mounted magnifier a fair bit, since it is quick to use, compared to setting up the camera (I need to dedicate space to that).

It’s the first day at BigBoxCo on the “Infrastructure” team. After working through the obligatory human resources activities, I received my contractor badge and made my way over to my new workspace. After meeting the team, I was told that we have a meeting with the “Recommendations’’ team this morning. My system access isn’t quite working yet, so hopefully IT will get that squared away while we’re in the meeting.

These supposed “dental inspection cameras” are easy to use and do not need any focussing because they have a relatively large depth of field. Unfortunately it is at the expense of picture quality. The reason is because the device has a very restricted aperture (i.e. a small pinhole) to get the depth of field, but the sensor is tiny and so the side effect is that the very low light level will result in a noisy image unless the PCB is flooded heavily with light. The built-in LEDs around the camera are not great; better to use an external lamp that can be aimed from an adjustable angle. Another issue is that the video image is compressed and doesn’t look great. In summary the dental USB camera approach to PCB inspection is barely usable.

About Camera, you can order generic chinese IP camera, same as mine, because they can be accessed by USB too. It shows up as UVC and you can use any available Webcam viewer for the same.

A device designed to pierce through darkness with intense luminosity. These spotlights are not just tools for illumination; they serve as beacons in ...

Machine vision/inspection cameras with a Gigabit Ethernet interface usually support a standard known as GigE Vision that allows them to interoperate with software from other vendors. They are also referred to as GigE cameras if they comply to the GigE Vision specification. For briefness I will refer to the physical interface as a GigEth interface.

Another view is shown below. The lamp (mains operated) was fitted with an LED lamp for a cool workspace. The Ethernet cable was held in place with Velcro. The assembly is just bolted to a piece of wood for now –it could be screwed directly onto a wooden table.

If you are buying a microscope for PCB inspection, you’ll need to check that you can get a good distance between the object to be observed and the lens assembly – otherwise larger components like electrolytic capacitors may get in the way when trying to read the markings on a small surface mount integrated circuit. This Meiji microscope allows for about 8cm working distance. It also has a built-in lamp with angle adjust capability so that the subject under test can be illuminated from the best direction for maximum contrast. The image is excellent and it is very easy to observe detail, this is the view looking down the eyepiece (10x magnification eyepiece, 2x magnification objective):

The angle clamp and mounting base was from Thorlabs. Not shown but important, there is a small thumbscrew-tightened collar underneath the angle clamp. It is needed to ensure the camera does not slam into the table accidentally whenever the clamp is adjusted.

After the meeting, I check my computer and it looks like I can finally log in. As I’m looking around, our principal engineer (PE) comes by and introduces himself. I tell him about the meeting with the Recommendations team, and he gives me a little more of the history behind the Recommendation service.

Power over Ethernet (PoE) capability is a nice-to-have feature because then just a single cable needs to connect to the camera and you get the benefits of energy efficiency and remote management and control. There are several PoE standards in existence. If you’re purchasing a new switch then ensure it has PoE+ capability at a minimum (also known as IEEE 802.3at-2009) and then the switch will be useful for powering many types of modern devices.

The duty cycle is typically calculated for 1 second of light on-off activity, as the target strobe rate is often provided in Hz or kHz. Duty Cycle Calculation:.

Ironman 4x4's LED Light Bars allow you to see further and draw less power, with state of the art, high intensity CREE LED's for maximum light spread.

Customized Mini Spotlight 1W 3W Focus Light Small ...

At the onset, this sounds to me like a graph problem. We have customers who log on to the site and buy products. Before that, when they look at a product or add it to the cart, we can show recommendations in the form of “Customers who bought X also bought Y.” The site has this today, in that the recommendations service does exactly this: It returns the top four additional products that are frequently purchased together.

In this way, the new recommendation data is constantly being kept up to date. Also, all of the infrastructure assets described above are located in the local data center. Therefore, the process of pulling product relationships from an order, sending them through a Pulsar topic and processing them into recommendations stored in Cassandra happens in less than a second. With this simple data model, Cassandra is capable of serving the requested recommendations in single-digit milliseconds.

I didn’t want to dedicate a DSLR for PCB inspection so I decided to use an industrial camera (used for things like machine vision, e.g. inspection of items on a conveyer belt in a factory). They are compact cube-shaped things, around the size of pool cue chalk. Often the lens is bigger than the camera.

The topic of PCB inspection magnifiers and cameras has come up many times on Element 14 forums. This blog post shows my experiences so far.

The PC network adapter in Windows was configured to enable jumbo frames up to 9014 bytes (to do this, navigate to the network adapter’s Network Connection Properties dialog and then select the Jumbo Packet property and set to 9014 bytes in the drop-down selection). Note that if you also use your PC to connect to other networks then you may need to disable this setting after you’ve finished using it with the camera.

2.5-Inch Outdoor Lighting Mounting Post with Cap and Stabilizing Wings for Landscape Lighting and Flood Lights, Black ... Stake (2-Pack).

The product recommendations are served to users on bigboxco.com via a microservice layer in the cloud. The microservice layer uses a local (cloud) data center deployment of Apache Cassandra to serve up the results.

Finally, the worst magnifier was one from a local store; it was intended for use as a clip-on book lamp; it was on sale and so I thought it was worth a try. However nothing was too clear through it! and I learnt that a book magnifier is almost useless for electronics work.

Another approach is to use an inspection loupe type of magnifier, which is designed for close-up work. One from Mitutoyo was used for the photograph below. It comes in two parts – the Mitutoyo handle with an incandescent bulbMitutoyo handle with an incandescent bulb and the Mitutoyo magnifier portionMitutoyo magnifier portion .

element14 is the first online community specifically for engineers. Connect with your peers and get expert answers to your questions.