Edge AI gateway combines Raspberry Pi CM4 with Coral Edge TPU

Edge AI gateway combines Raspberry Pi CM4 with Coral Edge TPU

Apr 2, 2021 — by Eric Brown

— 337 views

Techbase’s “iModGATE-AI” gateway for predictive maintenance is built around the RPi CM4 and a Coral Edge TPU AI module and offers up to 24-bit analog inputs and an advanced 9-axis IMU.

Techbase has announced an iModGATE-AI edge AI gateway that combines the Raspberry Pi Compute Module 4 or earlier RPi CM3+ with Google’s M.2 Accelerator equipped with its Coral Edge TPU. This is the same up to 4-TOPS at 0.5W, 2-TOPS per Watt AI chip found on the Coral Dev Board, Coral Dev Board Mini, LGA-based Coral Accelerator Module, and other Coral products.

iModGATE-AI and specs for iModGATE models including the ModBerry 500 CM4-derived iModM500 that it is based on
(click images to enlarge)

The iModGATE-AI can be programmed “to alert with certain rules, to prevent failures and enforce rapid maintenance,” says the company. “The Coral AI module and machine learning introduced to the end-point of the installation can minimize downtime and repair costs of machines, therefore maximizing the performance.” Applications include HVAC, wind farm monitoring, smart building & cities, conveyor systems, and robotics.

The iModGATE-AI is based on an earlier iModGATE design, which has yet to reach Techbase’s IIO+ Industrial IoT Shop where the iModGATE-AI is available for pre-order with 2-4 week delivery. An iModGATE/eModGATE product brief (PDF) suggests that the various listed iModGATE gateways are pre-configured systems based on its ModBerry line of mostly Raspberry Pi Compute Module based automation controllers.

The iModGATE systems run Techbase’s Ubuntu- and cloud-based iMod, iModCloud, and iModWizard Software-as-a-Service (SaaS) applications. The iModCloud platform provides file management, notifications, and custom-based actions, among other features. Compatible software runs on Techbase’s eModGATE systems, which are based on its ESP32-based controllers such as the ModBerry 500 with ESP32.

The iModGATE-AI shopping page links to a datasheet for the iModX500, which is the iMod version of the Raspberry Pi CM3+ based ModBerry 500, which was followed last October by the RPi CM4-based ModBerry 500 CM4. Long story short, the iModGATE-AI appears to be built around a choice of the CM3+ based iModX500 or CM4-based iModM500, which is listed in the iModGATE/eModGATE product brief. The latter is a specially configured and iMod-equipped variant of the highly modular ModBerry 500 CM4.

ModBerry 500 CM4 (left) and configuration chart for major SKUs
(click images to enlarge)

Key additions to the AI model along with the Edge TPU include an undisclosed number of “advanced” analog inputs with up to 24-bit resolution. (The iModM500 supplies 4x to 28x analog inputs.)

The analog inputs support a wide variety of market sensors, including the featured Botland Tensometric Beam NA1 shown in the image below. Tensometric beams are used for strain-gauge sensing in devices such as weight scales.

Optional Botland Tensometric Beam NA1 (left) and standard InvenSense ICM-20948 9-axis IMU
(click image to enlarge)

Another key addition is an InvenSense ICM-20948 9-axis MotionTracking module, which is not listed in the core specs for the iModM500 or ModBerry 500 CM4. This DMP (Digital Motion Processor) not only acts as a motion sensor but offers some built-in smarts that offload traffic processing algorithm calculations from the CM4 and “serve as a resource for predictive maintenance of the system,” says Techbase.

The 9-axis MotionTracking module includes a 3-axis gyroscope and 3-axis accelerometer, both with programmable FSR, as well as a 3-axis compass (magnetometer). The gyro has a programmable scale range of ±250 ~ ± 2000 dps and the accelerometer a range of ±2g ~ ±16g, offering a level of sensitivity that “reduces the requirements for calibration of the production line,” says Techbase.

The iModGATE-AI is further equipped with GbE, USB 2.0, and HDMI ports. There is also a 2×16 pin block terminal for various I/Os and an RTC with battery. The system defaults to an RPi CM4 with 32GB eMMC.

The image also shows a terminal block power input and reset button and a SIM card slot. This suggests it is based on the ModBerry 500 CM4 STD model, which adds a mini-PCIe slot among other features. Other likely features may be found in the iModM500 column of the table at top, as well as our ModBerry 500 CM4 report.

In February, Techbase announced a ClusBerry for Home desktop cluster system with 2x to 8x Raspberry Pi CM4 modules. The system offers modular configurations including serial I/O, wireless, AI, NAS, USB 3.0 hub, 2.5GbE, and SuperCap.

Further information

The iModGATE-AI is available for pre-order, with shipments in 2-4 weeks. (Pricing info requires “registration/login, wait/chat for authorization.”) More information may be found on the iModGATE-AI product page.